Examining Biological Motion as a Potential Factor in E-Scooter Conspicuity and Safety

Abstract

Background

E-scooter injuries have risen in recent years and riders report a relatively high prevalence of accidents. Collisions with motor vehicles pose a high risk to e-scooter users. E-scooter riders move fast relative to runners but lack movement of limbs that present aspects of biological motion to drivers, which may diminish conspicuity.

Method

Two experiments measured participants’ detection of point light representations beneath masking visual noise. Study 1 presented a runner, e-scooter rider, and rectangular object. Study 2 modified the e-scooter stimuli to remove motion sway and added alternative e-scooter presentations, one with moving lights consistent with biological motion and the other with the same motion in reverse, inconsistent with biological motion.

Results

Study 1 found a main effect of figure type, with the runner resulting in superior detection, recognition, and response time compared to the e-scooter rider, which performed better than the object. Study 2 found better perception performance for the runner, including better detection compared to the reverse motion e-scooter.

Conclusion

Findings suggest that reduced biological motion produced by e-scooter users slightly worsens and slows their detection by other road users and indicates an advantage for the perception of human body configurations. Any inclusion of apparent motion to improve detection, especially near the ground, should be consistent with biological motion.

Application

Visual display alterations (e.g., lighting) to introduce apparent motion that mimics biological movements or is consistent with biological motion may potentially confer a detection advantage over other movement patterns.

Keywords

perception-action bicycle and pedestrian safety reaction time experimental design surface transportation

Introduction

Since the emergence of electric scooter (e-scooter) sharing companies in 2017, e-scooter related injuries requiring emergency department attention have dramatically increased. Estimated emergency room visits for e-scooter related injuries increased from 4881 in 2014 to 26,628 in 2019 (Farley et al., 2020), and the US Bureau of Transportation Statistics (2024) reports that the number of US cities served by electronic scooter (e-scooter) systems has increased by 47% from 2020 to 2023, leading to an increased opportunity for crashes and injury. E-scooter riders report a relatively high prevalence of crashes, with 26.7% reporting a crash event and 15.2% reporting an injury (Tian et al., 2022). Rix and colleagues (2021) examined Texas traffic data and estimated e-scooters resulted in 180 injuries per million vehicle miles traveled (MVMT) compared to an estimated 0.9 injuries per MVMT for motor vehicles. These figures only account for injuries requiring emergency room intervention. Furthermore, bicyclists appear to have a slightly higher overall total and injury crash rate than pedestrians when accounting for volume and exposure on lower speed roads (Amoh-Gyimah et al., 2016; Beck et al., 2007; Siddiqui et al., 2012), but e-scooter riders report a higher relative rate of crash incidents compared to bicyclists and pedestrians (Sanders & Nelson, 2023), although clearly establishing the relative prevalence of crash rates by modality is not yet possible given the current quality of crash data.

Prospective and retrospective analyses of e-scooter related injuries and crashes have been conducted through patient interviews and reviews of patient health records (Mehranfar & Jones, 2024). At the beginning of this project, the authors examined the available research literature at the time, and found that the most prominent injury mechanisms were a loss of balance or falling off the scooter, collisions with a moving vehicle, hazardous roadway surface conditions, and collisions with another object (see Table 1), mirroring findings by a more recent scoping review of e-scooter safety (Janikian et al., 2024). Many of the reviewed studies were completed retrospectively, therefore some of the fall or loss of balance injuries likely resulted from more distal root causes, such as avoiding a motor vehicle, although infrastructure also likely played a critical role (White et al., 2023).

Table 1.

Top Three Injury Mechanisms for e-Scooter Injury Study.

Reference	Study Location	Sample Size	Most Common Mechanism of Injury	2nd	3rd
Austin Public Health (2019)	Austin, TX	192	Surface conditions (50%)	Excessive scooter speed (37%)	Collision with moto r vehicle (16%)
Bekhit et al. (2020)	Auckland, New Zealand	246	Balance/isolated fall (over 90%)	Collision with moto r vehicle (1.6%)	N/A
Blomberg et al. (2019)	Copenhagen, Denmark	112	Balance/isolated fall (86.6%)	Collision with moto r vehicle (8.9%)	Collision with stationary object (4.5%)
Bloom et al. (2021)	Los Angeles, CA	248	Balance/isolated fall (49%)	Collision with motor vehicle (14%)	Surface conditions (10%)
Brownson et al. (2019)	Auckland, New Zealand	180	Balance/isolated fall (91.7%)	Scooter malfunction (3.3%)	Collision with motor vehicle (2.8%)
Cicchino et al. (2021)	Washington D.C.	105	Balance/isolated fall (27%)	Surface conditions (25%)	Navigating road infrastructure (16%)
Dhillon et al. (2020)	Southern CA	87	Collision with moto r vehicle (57.5%)	Balance/isolated fall (23%)	Collision with stationary object (11.5%)
Liew et al. (2020)	Singapore	36	Balance/isolated fall (38.9%)	Collision with moto r vehicle (27.8%)	Collision with stationary object (16.7%)
Portland Bureau of Transportation (2018)	Portland, OR	176	Balance/isolated fall (83%)	Collision with moto r vehicle (12.5%)	Collision with pedestrian (1.7%)
Stormann et al. (2020)	Frankfurt, Germany	76	Balance/isolated fall (92.1%)	Collision w/ car (6.6%)	Collision w/ forklift truck (1.3%)
Trivedi et al. (2019)	Southern CA	228	Balance/isolated fall (80.2%)	Collision with stationary object (11%)	Collision with motor vehicle (8.8%)
Yang et al. (2020)	United States (News reports)	179	Collision with motor vehicle (∼65%)	Balance/isolated fall (∼30%)	Collision with pedestrian (∼5%)

Note. Bold font—fall; italic font—hazardous surface feature; both bold and italic font—involved a moving vehicle.

E-scooter conflicts with vehicles are a significant cause of injury. Shah and colleagues (2021) analyzed e-scooter and bicycle crashes caused by conflict with motor vehicles over a two-year period and found that most e-scooter crashes occurred at an intersection or at the junction of a driveway and roadway. The majority of e-scooter crashes occurred when the driver of the motor vehicle was driving straight or turning right, and the e-scooter was moving from the right of the motor vehicle or moving in the same direction as the motor vehicle. Sievert and colleagues (2023) surveyed e-scooter riders about perceived safety risks and found that uneven roadway surfaces and collision with a moving motor vehicle were of the greatest concern to e-scooter riders. Additional factors may increase the odds and severity of scooter crashes involving motor vehicles. For example, e-scooter riders were less likely to wear a helmet than cyclists and were more likely to ride under the influence of alcohol (James et al., 2023; Leyendecker et al., 2023).

Biological Motion

One factor that may increase the risk of e-scooter crashes with motor vehicles is the relative absence of biological motion produced by e-scooter users. Pedestrians and cyclists generate a range of recognizable biological motion cues through the movement of their arms, legs, shoulders, and hips, although bicyclists may present less overall biological motion than pedestrians. E-scooters require minimal physical input from their riders, and the reduction of biological motion may lead to subsequent increased difficulty of visual detection by drivers.

The classic paradigm in biological motion research employs point light (PL) displays, during which participants are shown footage of actors walking with light points visible at their head and joints, but with no other parts of the body visible, with the resulting pattern consistently identified by participants as a walking person (Johansson, 1973). Studies considering biological motion with PL displays have found that humans can detect a variety of different actor characteristics such as type of movement, emotions, and sex (Dittrich, 1993; Dittrich et al., 1996; Kozlowski & Cutting, 1977). Observers are also capable of making identifications when some light points are removed, and in the presence of noise and distractors (Cutting et al., 1988; Mather et al., 1992). PL dots representing the extremities of the body were most crucial to recognition and complex trajectories of motion in the extremities were identifiable as belonging to a walking human (Mather et al., 1992). Cutting and colleagues (1988) evaluated various types of noise masks and their effectiveness at obscuring PL walkers from detection of implied motion direction and found that only complex and significant noise masking could substantially reduce the direction detection of PL figures from ceiling performance when the figures were presented up to 400 milliseconds or longer.

Biological motion indicators can improve pedestrian and bicyclist conspicuity, especially at night (Wood, 2023). Owens and colleagues (1994) were among the first researchers to examine how biological motion impacts the detectability of pedestrians at night. Retroreflective markings highlighting biological motion at the extremities resulted in significantly longer time-to-collision detections of pedestrians than a retroreflective vest (Owens et al., 1994). Multiple studies have demonstrated the beneficial effects of highlighting biological motion in pedestrians at night (Balk et al., 2008; Black et al., 2021, 2023; Luoma et al., 1996; Wood et al., 2014, 2017). The use of biological motion indicators is unsurprisingly more effective when the pedestrian is in motion relative to standing still (Mian & Caird, 2018), and provides a detection advantage in environments with significant visual clutter (Tyrrell et al., 2009). For bicyclists, participants reported that when a taillight or florescent marking was placed on the extremities (i.e., ankles and legs) of a bicyclist, the bicyclist was more conspicuous at significantly greater distances, with fewer recognition failures, relative to placements at the seat or the helmet (Edewaard et al., 2020), torso (Wood et al., 2022) or the posterior of the bicyclist (Blomberg et al., 1986).

E-scooter users typically stand upright and motionless with one foot ahead of the other on the scooter platform, with both hands on the handlebars, leading to minimal presentation of biological motion (see Figure 1). Although higher relative speeds are likely a major factor, the reduction of biological motion cues may contribute to findings of a higher rate of reported e-scooter crashes relative to pedestrian crashes (Karpinski et al., 2023; Sanders & Nelson, 2023). Bicyclists only present partial biological motion through their legs, and at some viewing angles (e.g., from the side) only one moving leg may be easily viewable. Same-side or ipsilateral limb movements are perceived better than random movement configurations, but not as well as different side or contralateral limb movements or whole-body movements (Pinto & Shiffrar, 1999). The reduced biological motion presented by bicyclists may potentially contribute to the higher rate of overall crashes for bicyclists relative to pedestrians, along with speed and exposure factors (Amoh-Gyimah et al., 2016; Beck et al., 2007; Siddiqui et al., 2012). Another consideration is the competition for visual attention in the driving environment. Scooter deployment is relatively recent so there is less time to develop well-established and learned visual cues. Consequently, drivers of vehicles may be less likely to have accurate situation awareness of the travel dynamics of those traveling on e-scooters.

Figure 1.

Still images of a pedestrian point light (PL) walker progressing through time/motion compared to a point-light e-scooter rider progressing through time/motion. These PL figures were not used in the following studies.

Biological motion predicts that the runner PL figure will be detected more accurately and quickly than a scooter or an object figure. There are other potential alternative explanations governing detection of stimuli which may be relevant in this context. The gestalt principles of proximity and common fate suggest that overall object motion with points moving together and independent relative to other stimuli in the visual field may lead to greater detectability of the object (Spillmann, 2012; Wertheimer, 1923/2012). This overall object motion hypothesis would suggest that the additional motion of the runner and any residual motion presented in the PL scooter figure would lead to greater detectability, relative to rigid objects. Another potential explanation is that the organization of the point lights for the e-scooter rider presents configural information consistent with the structure of the human body, and this configural form detection plays a major role in the perception of biological motion (Lange & Lappe, 2006). Research with PL displays indicates that people exploit human configural arrangements without movement (Pinto & Shiffrar, 1999). Applied research examining the detectability of pedestrians at night by applying retroreflective markings on the pedestrians found that the best detection rates were found with the full biological motion condition with multiple limb segments (e.g., arm, leg, and torso) made visible and in walking motion, but detection performance was also superior for that marking condition when standing (i.e., not in motion), relative to the other nonmotion conditions (Balk et al., 2008). If the human body configuration has an advantage in detection, this would predict superior detection advantage for the runner and scooter PL figures relative to the object, irrespective of motion.

The current study explored whether the relative absence of biological motion cues produced by e-scooter users makes them more difficult to detect visually by using PL displays of a runner, an e-scooter rider, and a control object in motion within a mask of distractor particles. The authors selected the approach of PL figures as this is the original format under which biological motion was examined in perception (Johansson, 1973) and allows for such examination under short spatial and temporal intervals without significantly involving other confounding higher-order visual processes (Mather et al., 1992). This helps us better examine whether biological motion is a plausible contributory factor in detection of e-scooters, at the cost of reducing real-world generalizability of the immediate findings. We hypothesized that the e-scooter PL figure would be more difficult for participants to visually detect, potentially indicating the reduction of biological motion as a contributor to crashes and injuries resulting from conflict between e-scooters and motor vehicles.

Study 1 Method

Participants

An experiment was conducted with 62 participants aged 18–60 years old (M = 35.76, SD = 13.22). A recruitment and online screening survey was distributed via flyers posted on a large public university campus in the Midwest, shared on social media accounts, and advertised on Craigslist. Prospective participants verbally reported the absence of epilepsy or other sensitivities to flashing lights. Demographic information was electronically collected from study participants in person, via a Qualtrics™ survey, including age, gender, education, and racial background. Of the 62 participants, 40 (64.5%) were male, 21 (33.9%) were female, and 1 (1.6%) was nonbinary. The mean age of participants was 36 years (SD = 13.1). Other survey materials included a driving history questionnaire, whether and how often participants had ridden e-scooters in the past. This research complied with the American Psychological Association Code of Ethics and was approved by the Institutional Review Board at the University of Minnesota (Study ID: STUDY00010556). Informed consent was obtained from each participant.

Study 1 Apparatus and Materials

Point Light Representations

Moving point light (PL) representations were displayed as video clips beneath a layer of dots (called noise or distractors) which move at various speeds and paths to make the recognition task more difficult, with parametric manipulation of noise levels for 40,000 (40k), 60,000 (60k), and 80,000 (80k) noise dots. The approximate density of distractor dots was 0.2068 (40k), 0.3102 (60k), or 0.4136 (80k) particles per cubic meter. Three target PL characters were used that represented (1) a running human, (2) a human riding an e-scooter, and (3) a cube-like shape (“object”), as well as a final condition (4) no object present (see Figure 2). Each figure was represented with 13 white points to keep the number of visual stimuli constant across figure conditions. The points of the runner were articulated to represent the motion of a human. The points of the scooter rider and object were fixed relative to each other. All video clips were displayed for 1 s, with a window of 1 additional second after display in which participants were still able to respond. The height (1.518 m) and width (.564 m) of the object approximated the maximal vertical and horizontal distances between the points of the runner (height = 1.518 m; width = .429 m) and the e-scooter rider (height = 1.648 m; width = .429 m). All objects were in motion and had a speed of 3.12 m per second and moved in one of four directions: left across the screen, right across the screen, “into the screen” or away from the observer, and “out of the screen” or towards the observer. The camera or viewing perspective was set at 2.822 m distance from the center where the PL figure crossed in left/right direction conditions, 2.540 m distance in toward conditions, and 3.104 m distance in away conditions. The position and articulation of the runner’s points were created using free-use motion capture data from Carnegie Mellon University, see author note and data availablity statement. The scooter rider figure was created by taking the human joint position in the CMU data and matching it to a Segway Ninebot Max e-scooter.

Figure 2.

Images of PL target characters: runner, e-scooter rider, and a box-like object (left to right). For visual clarity in this paper, duplicate images of the runner and e-scooter rider are provided with lines to indicate the connected joints. Participants did not see the lines.

Testing was conducted in a laboratory environment with overhead fluorescent lighting conditions consistent across participants. Participants were positioned approximately 90 cm away from the screen, with the PL video having an on-screen resolution of 1620 × 1080 with physical dimensions of 15.5 cm × 10.0 cm. The screen used was a Dell P4317Q with a resolution of 3840 × 2160. Using JavaScript, a set of up to 80,000 randomly positioned distractor points was generated about the center of the presented environment, which consisted of a horizontal radius of 45.2 m and a vertical radius of 22.6 m (see Figure 3). Each subset of 1000 points out of the 80k (k = 1000) randomly rotated about all three axis (x-, y-, z-axis). A 32 × 32 pixel PNG of a white circle was attached to each point. During a trial all points were in motion at a speed of 1.7 m per second. Motion of the sets was evenly distributed such that one-quarter of the sets moved in each direction (toward, away, left, and right). For the monitor in this study, the pixel pitch was 0.2451 mm in diameter. The apparent visual size of each revolving dot would be about 31.86 mm in diameter for the dots “nearest” to the viewer, and 1.23 mm in diameter for the dots “farthest” from the viewer, with a viewing distance of 900 mm from the viewer to the monitor.

Figure 3.

Representation of visual noise radius in a sphere (∼45 m diameter) along with visual noise direction around the moving point light target (i.e., stick figure). The participant’s perspective (white arrows) is within the visual noise sphere, approximately 2.5–3.1 m from the center depending on direction condition, viewing the moving PL figure crossing the center of the sphere moving either directly toward or away from the participant or moving left or right across the participant’s view.

Procedure

After providing informed consent and answering the questionnaires, participants received an explanation as to the format of each trial. Each trial displayed a video clip containing the mask and point light figure (if present). If the participant saw something beneath the noise, they responded with a space bar press. Reaction time to press the spacebar was measured. The participant’s task was to identify the presence or absence of a PL figure and the direction of motion of the PL figure if one was present, that is, runner, e-scooter rider, object, no figure, or unsure via key press (R, S, O, N, and X, respectively). Finally, if a PL target was present, to identify the direction the point light representation was moving via arrow key press, that is, left across the screen (←), right across the screen (→), “into the screen” or away from the observer (↑), and “out of the screen” or towards the observer (↓). Participants were also able to answer “Unsure” to type of figure and directionality, providing better granularity to the resulting data by ensuring visual detection of motion could still be identified as having taken place—even if the participant was not confident in what they saw. See Figure 4 for a demonstration of two enhanced frames of the runner and e-scooter rider in motion (not as depicted in the study).

Figure 4.

Frame captures of point light runner (left) and e-scooter rider (right) in motion with white point lights within white point distractor lights (noise). The PL are indicated here with grey dots for contrast (white in the experiment) and connected in lines for the reader. The lines were not present in the experiment.

Experimental Design and Trials

The experiment was a 4 (Character) × 4 (Direction) × 3 (Noise Level) factorial repeated measures design. The independent variables of the study were the type of PL figure (runner, e-scooter rider, object, and no stimuli) and the noise distractor level. Levels of noise were blocked in fixed ascending order (40k → 60k → 80k particles) (k = 1000). Participants were exposed to the 12 unique stimuli/direction combinations with a point light figure presented twice per distractor level, and the no stimuli condition (mask only) presented 24 times per distractor level (50% of total trials). See Table 2 for a summary. The number of experimental trials per distractor level was 48. The total number of experimental trials across the 3 distractor levels was 144. Dependent variables were correct detection (whether a figure was shown), recognition count, and response time.

Table 2.

Summary of Study 1 Experiment Design.

Stimuli (% of Trials)	Direction Representation Traveling				Number of Point Lights in Representation
	Left	Right	Towards	Away
	Shorthand denotation for each unique trial
Runner (16.67%)	RL	RR	RT	RA	13
E-scooter (16.67%)	EL	ER	ET	EA	13
Object (16.67%)	OL	OR	OT	OA	13
No stimuli (50%)	-	-	-	-	N/A

Practice Trials

Participants were shown a round of practice trials to familiarize them with the task and the keyboard controls for responding during the task before immediately advancing to the 144 experimental trials. The practice rounds consisted of 25 different trials.

• Practice Set 1: 12 trials, each figure, in each direction, with no noise particles.

• Practice Set 2: 1 trial showing only the noise with 40k particles.

• Practice Set 3: 12 trials displaying each figure, in each direction, with 40k particles.

Participants were given instructions with the practice trials to help them understand the task. For responding to the figures, the participants were told: “When you think you see the figure, you’ll tap space […] There is a timer that goes fairly quickly when each figure starts to display, so responding quickly with space is important as well.” After being shown Practice Set 2 (noise trial), participants were told: “So that is the first example of the noise or clutter that is going to be obscuring the figures. You are still looking for the same figures that you saw in the practice, just moving behind this cloud of dots. Now that the noise is being presented, there is the option that no figure will be shown—it may be only noise.”

Measures and Analyses

The following analyses consider the effects of the independent repeated measures variables of character (runner, scooter, and object) and distractor (40k, 60k, and 80k) on detection accuracy, recognition count, and response time. For analyses, this resulted in a 3 × 3 repeated measures design. The direction the point light figure traveled with respect to the viewer (left, right, towards, and away) were averaged together across these analyses. The no-stimuli condition was used to determine detection accuracy but was not otherwise considered in these analyses. For target detection, the analysis examined the signal detection theory measure of A’. A’ can better account for ceiling and floor performance (Stanislaw & Todorov, 1999). False alarm rates were calculated for nonevent trials for each level distractor (40k, 60k, and 80k). For recognition count, the sum of correct identifications for a given character in a particle condition was calculated, with higher counts indicating greater recognition accuracy with a total possible count of 8 per cell. For the response time when individuals indicated that they detected a target, we used nontransformed reaction times beginning with trial initiation and ending with the time the participant pressed the response button. Trials lasted for 2 s, resulting in a range of values from 0 to 2 s. Repeated measures ANOVAs were conducted for the analysis of each dependent variable.

Study 1 Results

The descriptive statistics are presented in Table 3. For A’, there was a significant Character by Distractor interaction, F (4, 160) = 2.543, p = .042, ω² = .004. Post-hoc analyses suggested this interaction resulted from a statistically significant difference for detection accuracy between the Runner and Object conditions for the 40k, 60k, and 80k distractor levels (Cohen’s d = .238, .331, .435, respectively; ps < .05). However, there was only a difference between the Runner and Scooter conditions at the 60k distractor level, t (40) = 3.335, p = .004 (Holm corrected), d = .277, and between the Scooter and Object conditions at the 80k distractor level, t (40) = 3.881, p = .001 (Holm corrected), d = .423. There was a main effect of Distractor, Huynh-Feldt corrected, F (1.571, 62.856) = 15.790, p < .001, ω² = .055. There was no difference in average detection accuracy for the 40k and 60k distractor conditions (p = .062, Holm corrected), but there was a difference between the 40k and 80k conditions, t (40) = 4.395, p < .001 (Holm corrected), d = .550, and the 60k and 80k conditions, t (33) = 4.178, p < .001 (Holm corrected), d = .403. There was a main effect of Character, F (2, 80) = 15.881, p < .001, ω² = .022. Detection accuracy for the Runner condition was significantly better than that of the Scooter condition, t (40) = 2.605, p < .013 (Holm corrected), d = .151, and the Object condition t (40) = 5.339, p < .001 (Holm corrected), d = .335. Detection accuracy for the Scooter condition was significantly different than the Object condition, t (40) = 3.183, p = .006 (Holm corrected), d = .184.

Table 3.

Study 1 Descriptive Statistics.

Distractor	Character	Recognition Count Average (SD)	A’ Average (SD)	Response Time (sec) Average (SD)
40k
	Runner	4.676 (2.982)	.771 (.266)	.999 (.149)
	Scooter	3.147 (2.271)	.729 (.272)	1.032 (.168)
	Object	2.588 (2.148)	.710 (.260)	1.083 (.172)
60k
	Runner	6.059 (1.922)	.751 (.245)	1.092 (.194)
	Scooter	1.912 (.1.798)	.680 (.259)	1.057 (.159)
	Object	1.206 (1.805)	.666 (.256)	1.132 (.218)
80k
	Runner	1.941 (1.969)	.634 (.254)	1.086 (.169)
	Scooter	1.206 (1.200)	.631 (.247)	1.103 (.171)
	Object	.647 (1.475)	.522 (.251)	1.159 (.205)

Note. k = 1000.

For recognition count, there was a significant Character by Distractor interaction, F (4, 132) = 24.032, p < .001, ω² = .190. Post-hoc analyses indicated that there were significant differences in recognition counts between the Runner and Scooter conditions for the 40k distractor level, t (33) = 3.482, p = .003 (Holm corrected), d = .761, and the 60k distractor level, t (33) = 9.767, p < .001 (Holm corrected), d = 2.061, but not for the 80k distractor level, t (33) = 2.321, p = .053 (Holm corrected), d = .366. The Runner and Object character conditions were significantly different for all Distractor conditions (ps < .05), and the Scooter and Object character conditions were not significantly different for all Distractor conditions (ps > .09). There was a main effect of Distractor, Huynh-Feldt corrected, F (1.769, 58.387) = 44.713, p < .001, ω² = .308. There was a significant difference for the 40k and 80k conditions, t (33) = 8.402, p < .001 (Holm corrected), d = 1.098, and the 60k and 80k conditions, t (33) = 9.475, p < .001 (Holm corrected), d = .893, but no significant difference between the 40k and 60k conditions, t (33) = 1.457, p = .154 (Holm corrected), d = .205. There was a main effect of Character, F (2, 66) = 35.667, p < .001, ω² = .338. There was no significant difference in recognition for Scooter and Object conditions, t (33) = 1.774, p = .085 (Holm corrected), d = .302, whereas there was a significant difference between the Runner and Scooter conditions, t (33) = 7.508, p < .001 (Holm corrected), d = 1.064, and the Runner and Object conditions, t (33) = 7.057, p < .001 (Holm corrected), d = 1.366.

For response time, there was no observed interaction between Distractor and Character (p = .547). There was a main effect of Distractor, Huynh-Feldt corrected, F (1.772, 60.263) = 5.191, p = .011, ω² = .041. There was a significant difference between average response times for the 40k and 60k conditions, t (34) = 2.786, p = .017 (Holm corrected), d = .315, and the 40k and 80k conditions, t (34) = 3.163, p = .010 (Holm corrected), d = .441, but not for the 60k and 80k conditions (p = .452, Holm corrected). There was a main effect of Character, F (2, 68) = 6.066, p = .004, ω² = .038. There was a significant difference in response times for the Runner and Object conditions, t (34) = 3.140, p = .010 (Holm corrected), d = .360, and the Scooter and Object conditions, t (34) = 2.620, p = .026 (Holm corrected), d = .331, but no difference between the Runner and Scooter conditions (p = .782).

Preliminary Discussion

The results of the first study indicated a significant effect of PL figure type, with the runner detected in noise more frequently than the e-scooter rider or object. Participants responded faster when shown the runner compared to the object. In general, participant performance in the e-scooter figure condition was worse than in the runner condition, but better than in the object condition. Although the results support the role of biological motion perception in explaining the differences between the PL runner condition and the other conditions, it is less clear why performance in the e-scooter condition was better than performance in the object condition. While the point lights of the figures had similar hue, luminance, and size relative to the distractor particles, the distance between the point lights in the e-scooter rider were closer in spacing relative to the object and had a very slight sway, like a slight lateral sway for a real e-scooter rider that is moving longitudinally on a path. The object figure moved with no sway. This suggests that there are potential effects of either biological motion, human body configuration, overall object motion, or some combination of those three factors. A second study was conducted to partially disentangle these factors.

The second study had the following changes. First, the PL e-scooter figure motion sway was removed. If increased overall object motion is a major contributor, we predicted that this would lead to comparatively worse performance for the scooter condition. Second, two new conditions were also added, presenting two alternate scooter conditions with the handlebars appearing and disappearing in a contralateral or ipsilateral flashing pattern relative to the bottom lights, and the bottom point lights on the figure along the deck of the scooter were now traveling in either a forward or backward motion. The forward motion and contralateral handlebar-footfall pattern is consistent with biological motion, as a footfall and opposite hand raise made by a runner moving in the same direction of motion as the PL scooter figure, whereas the backward or reverse motion and ipsilateral handlebar-footfall pattern is not consistent with biological motion generated by a runner moving in the same direction. These two areas of handlebars and e-scooter deck were chosen to be in motion and flashing to best integrate the scooter PL figure with similar levels of motion either consistent with or inconsistent with biological motion, because those areas of an actual e-scooter (e.g., along the deck sides) could be outfitted with a lighting system to provide additional visual salience to increase detectability of scooters in a multimodal travel environment. If there is a general perceptual advantage for biological motion, the prediction was better performance for the scooter condition consistent with biological motion relative to the reverse motion scooter condition inconsistent with biological motion.

Study 2 Method

Participants

An experiment was conducted with 66 participants, aged 18–67 years (M = 31.83, SD = 12.41). Prospective participants were screened for the absence of epilepsy or other sensitivity to flashing lights. Of the 66 participants, 41 (62.1%) were male, 24 (36.4%) were female, and 1 (1.5%) was nonbinary. Similar study recruitment methods were employed as in Study 1. Eleven (11) participants participated in both studies. There was minimal concern with performance improvement across studies, as the studies took place approximately 10–12 months apart. To control any potential effects of prior learning, the performance of the 11 participants was examined across studies, described in the Study 2 results section. This research complied with the American Psychological Association Code of Ethics and was approved by the Institutional Review Board at the University of Minnesota (Study ID: STUDY00021713). Informed consent was obtained from each participant.

Apparatus and Materials

Point Light Representations

Moving point light (PL) representations were displayed as video clips beneath a layer of dots (called noise) that were also moving at various speeds and paths to make the recognition task more difficult, with parametric manipulation of noise levels. 5 target PL characters were used that represented (1) a running human, (2) a human riding an e-scooter, (3) a human riding an e-scooter with a pattern of moving/blinking lights meant to mimic biological motion (see Figure 5), (4) a human riding an e-scooter with an alternate lighting pattern (the motion pattern reversed, see Figure 6), and (5) a cube-like shape (“object”), as well as a final condition (6) no object present. Each character was represented with 13 white points. The points of the runner were articulated to represent the motion of a human. The points of the scooter riders and object were fixed relative to each other, except for the moving/blinking lights for two of the e-scooter variants. All other parameters and procedures were consistent with the design of Study 1. Further examples of the stimuli can be found in the supplemental materials.

Figure 5.

Frame captures of human riding e-scooter with light pattern consistent with biological motion for side profile (left three images) and forward-facing profile (right two images) without noise. Light position changes through time (forward motion footfall and contralateral hand and foot indication) are highlighted with arrows and figure point lights are connected in lines for demonstration.

Figure 6.

Frame captures of human riding e-scooter with reversed light pattern inconsistent with biological motion for side profile (left three images) and forward-facing profile (right two images) without noise. Light position changes through time (backward motion for footfall and ipsilateral hand and foot indication) are highlighted with arrows and figure point lights are connected in lines for demonstration.

Measures and Analyses

For the second study, analyses were performed on the same measures as the first study, with repeated measures ANOVAs. The directions the PL figure traveled with respect to the viewer were averaged together across these analyses. A preliminary analysis examined the 11 participants that participated in both studies, examining Study 1 and Study 2 performance for each dependent variable to determine if there was an improvement for any of the shared character conditions. Afterward, the first analysis replicated and extended the analysis in the first study with 5 levels of Character (Runner, Scooter, Object, Scooter Biological Motion Consistent, and Scooter Reverse Motion) and 3 levels of Distractor (40k, 60k, and 80k).

Study 2 Results

The descriptive statistics for Study 2 are presented in Table 4. The preliminary analyses with paired-sample t-tests for the 11 shared study participants found no effect of the character variables between studies for response time (ps > .19, Bonferroni corrected). For A’, there was no effect of the character variable between studies (ps > .66, Bonferroni corrected). A marginal difference between studies was found for recognition count for the Runner character (t (6) = 2.242, p = .066, Bonferroni corrected, d = .776), with an apparent decline in performance for recognition count from Study 1 (M = 4.33) to Study 2 (M = 2.91), with no trend apparent for the other character conditions (ps > .17). To ensure that the assessment of character recognition was not influenced by exposure across studies, these eleven participants were excluded from the Study 2 analysis of the recognition counts dependent variable.

Table 4.

Study 2 Descriptive Statistics.

Distractor	Character	Recognition Count Average (SD)	A’ Average (SD)	Response Time (sec) Average (SD)
40k
	Runner	4.387 (2.765)	.772 (.241)	.990 (.178)
	Scooter	3.355 (2.916)	.780 (.215)	.994 (.157)
	Object	2.452 (2.293)	.762 (.218)	1.001 (.154)
	Scooter B	2.839 (2.899)	.775 (.180)	1.009 (.176)
	Scooter R	2.871 (2.871)	.754 (.225)	.992 (.147)
60k
	Runner	2.774 (2.539)	.781 (.207)	1.052 (.221)
	Scooter	2.258 (2.516)	.739 (.220)	1.036 (.192)
	Object	1.387 (1.892)	.747 (.176)	1.041 (.183)
	Scooter B	2.419 (2.592)	.743 (.199)	1.046 (.196)
	Scooter R	2.548 (2.755)	.730 (.202)	1.050 (.177)
80k
	Runner	1.548 (1.947)	.704 (.182)	1.126 (.229)
	Scooter	1.677 (2.271)	.692 (.201)	1.121 (.239)
	Object	.677 (1.222)	.201 (.201)	1.137 (.251)
	Scooter B	1.226 (1.978)	.620 (.204)	1.159 (.210)
	Scooter R	1.323 (1.222)	.620 (.230)	1.107 (.239)

Note. k = 1000. The recognition count averages are from participants who participated in Study 2 only.

For A’, there was no Character by Distractor interaction, F (8,440) = 1.582, p = .128, ω² = .002. There was a main effect of Distractor, Huynh-Feldt corrected, F (1.563, 85.990) = 17.454, p < .001, ω² = .059. The 80k distractor condition was significantly different than the 40k distractor condition, t (55) = 4.544, p < .001 (Holm corrected), d = .525, and the 60k distractor condition, t (55) = 4.635, p < .001 (Holm corrected), d = .424. The 40k and 60k distractor conditions were not significantly different, t (55) = 1.433, p = .157 (Holm corrected), d = .100. There was a main effect of Character, Huynh-Feldt corrected, F (3.558, 195.711) = 4.541, p = .002, ω² = .008. The Runner condition had a significantly higher average detection accuracy compared to the Scooter Reverse Motion condition, t (55) = 3.427, p = .012 (Holm corrected), d = .244. The Scooter condition and the Scooter Reverse Motion conditions were marginal but not significantly different, t (55) = 2.752, p = .072 (Holm corrected), d = .171. The other conditions were not significantly different (ps > .12).

For recognition count, the following analysis excluded the 11 participants that were in both studies. There was a significant Character by Distractor interaction, Huynh-Feldt corrected, F (6.572, 197.167) = 2.445, p = .023, ω² = .009. Post-hoc analyses indicated that for the 40k distractor level, there were significant differences in recognition count between the Runner and Object conditions, t (30) = 3.420, p = .016 (Holm corrected), d = .808, the Runner and the Scooter Biological Motion Consistent conditions, t (30) = 3.625, p = .011 (Holm corrected), d = .646, and the Runner and the Scooter Reverse Motion condition, t (30) = 3.223, p = .024 (Holm corrected), d = .633. For the 40k condition, the Runner and Scooter conditions were not significantly different for recognition count, t (30) = 2.513, p = .123 (Holm corrected), d = .431, and the other conditions were not significantly different (ps > .57). For the 60k and 80k distractor levels, there were no significant differences in recognition count between character conditions (ps > .128). There was a main effect of Distractor, Huynh-Feldt corrected, F (1.625, 48.738) = 35.114, p < .001, ω² = .135. There was a significant difference for the 40k and 80k conditions, t (30) = 7.004, p < .001 (Holm corrected), d = .789, the 60k and 80k conditions, t (30) = 6.125, p < .001 (Holm corrected), d = .412, and the 40k and 60k conditions, t (30) = 3.890, p < .001 (Holm corrected), d = .377. There was a main effect of Character, Huynh-Feldt corrected, F (2.538, 76.138) = 5.009, p = .005, ω² = .035. There was a significant difference in recognition for Runner and Object conditions, t (30) = 3.381, p = .020 (Holm corrected), d = .583, whereas there were no significant differences between the other conditions (ps > .14).

For response time, there was no observed interaction between Distractor and Character, F (8, 368) = .570, p = .803, ω² = .000. There was a main effect of Distractor, Huynh-Feldt corrected, F (1.570, 72.209) = 25.045, p < .001, ω² = .085. There was a significant difference for the 40k and 60k conditions, t (46) = 3.277, p = .002 (Holm corrected), d = .242, the 40k and 80k conditions, t (46) = 5.645, p < .001 (Holm corrected), d = .668, and the 60k and 80k conditions, t (46) = 4.778, p < .001 (Holm corrected), d = .426. There was a no effect of Character F (4, 184) = 1.032, p = .392, ω² < .001). See Figure 7 for a comparison of the two studies for each dependent variable.

Figure 7.

Estimated marginal means for the Character condition for the first and second studies. Scooter B is the scooter condition with movement consistent with biological motion. Scooter R is the scooter condition with reverse movement, inconsistent with biological motion. Response time is in seconds. Error bars represent standard error (SE).

General Discussion

The emergence of e-scooter sharing companies in 2017 contributed to a pronounced increase in e-scooter-related injuries requiring emergency department attention, with counts of e-scooter related emergency room visits more than quintupling from 2014 to 2019 (Farley et al., 2020). E-scooters have a relatively high prevalence of injuries among users compared with other modes of transit (Tian et al., 2022), exhibiting significantly more injuries per vehicle miles traveled than motor vehicles (Rix et al., 2021). A major concern of e-scooter riders is collision with motor vehicles (Sievert et al., 2023). A significant body of work exists on biological motion and the influence it can have on improving safety by enhancing the detectability of pedestrians and cyclists by motor vehicle drivers (Owens et al., 1994; Edewaard et al., 2020). However, there is a lack of research regarding e-scooters and the effects the reduction of biological motion and other factors may have on the perception and conspicuity of e-scooter riders. This study examines this question and evaluates the detection, recognition, and response time of participants shown point light (PL) representations of runners and e-scooter riders.

For Study 1, the runner was detected in noise more frequently, detected more quickly, and more accurately identified than the e-scooter rider or object. In general, participant performance in the e-scooter figure condition was worse than in the runner condition, but better than in the object condition. For Study 2, the runner had superior detection performance compared to the reverse motion scooter condition, and the runner was more readily identified relative to the object and the two scooter motion conditions in specific distractor conditions. Altogether, these findings are partly consistent with the hypothesis that the diminished availability of biological motion cues produced by e-scooter riders may make it more difficult for people to perceive riders.

Research Implications

Although the results partially support the role of biological motion perception in explaining the differences between the PL runner condition and the other conditions, it is less clear why performance in the e-scooter condition was better than performance in the object condition in the first study. There was a significant difference in detection between the runner and scooter in Study 1 for A’, which was not replicated in Study 2. Removing the sway may have slightly improved the detectability of the scooter (Figure 7), suggesting that overall object motion does not account for the results in Study 1. Furthermore, introducing motion with the alternative scooter figures in Study 2 did not improve the detectability of those figures relative to the original scooter figure without internal motion. The human body configuration explanation, which argues for a configural form mechanism or form pathway for the biological motion advantage (Lange & Lappe, 2006), remains relatively convincing, as the scooter PL figure remains more detectable than the object in Study 1 and there were no observed differences between the modified scooter PL figure and the runner in Study 2.

However, this configural form or human body explanation cannot solely account for the detection advantage. Given that the human body is represented in all three scooter PL figures in Study 2, we would expect roughly equivalent performance between the three scooter conditions relative to the runner, which is not what was found. Introducing motion for part of the figure significantly reduced the detectability of the reverse motion scooter figure. Multiple mechanisms contribute to the perceptual advantages of biological motion, which is reasonable given the relatively wide distribution of brain regions involved (Chang et al., 2018). A configural form perception pathway via global analyses (Lange & Lappe, 2006; Shiffrar et al., 1997) and a local motion pathway for detecting biological motion exists (Bromfield & Gold, 2017; Chang & Troje, 2009; Troje & Westhoff, 2006). The local motion pathway for biological motion is most sensitive to locomoting limbs, specifically feet in the case of human locomotion (Bromfield & Gold, 2017). This arguably is an efficient way to identify biological organisms that may be a threat or may be beneficial (e.g., social partners) (Troje & Chang, 2023). Another possibility for the reverse motion disadvantage is the integration of conflicting configural form, temporal pattern, and local motion information in biological motion perception, as conflicting local and global cues can lead to worse performance (Thurman & Lu, 2013).

Practical implications

E-scooters can travel at speeds up to 18 mph and the relative absence of biological motion cues may contribute to a greater rate of collisions due to drivers not detecting the rider in time. Furthermore, while cities differ on their approach for managing e-scooter use, some municipalities have introduced regulation that e-scooter riders must use the street instead of the sidewalk to improve pedestrian safety and comfort (Jafari & Liu, 2024; Orozco-Fontalvo et al., 2023), which may contribute to a much greater proportion of motor vehicle-involved crashes and severe injuries requiring hospital treatment (Neuroth et al., 2022). Drivers must rely on other motion and distance cues which may be absent or impoverished in certain contexts (e.g., low light, precipitation, fatigue, and complex environments) increasing crash and serious injury risks to e-scooter riders.

Biological motion remains a viable explanation for the superior detectability of the runner character. One design implication of these findings is that mimicking biological motion through visual displays could be a useful way to better capture attention. Study 2 provided a preliminary test of biological motion visual displays with the alternative scooter figures. Introducing motion into part of the scooter figure in the distractor-laden PL display served to reduce overall detectability of the scooter figures, but the use of motion consistent with biological motion partially mitigated this reduction in detectability, given a significant difference between the runner figure and the reverse motion scooter figure, with no statistically significant difference between the biological motion consistent scooter figure with apparent motion consistent with biological motion and the runner figure in detectability. However, comparing roughly equivalent conditions that mostly differed on whether they possessed qualities consistent with biological motion (e.g., runner vs. nonsway scooter; biological motion consistent scooter vs. reverse motion scooter) found the range of effect sizes were all fairly small (d = .073 and d = .084, respectively, for A’).

For design recommendations, recent research has demonstrated that improving e-scooter visibility with by increasing color contrast improves e-scooter detection time by drivers (Komova et al., 2024); therefore, the presentation of organized flashing lights would likely have a large effect on detectability and conspicuity (Rumar, 1990). Motion onset is known to capture attention (Smith & Abrams, 2018), therefore using lighting in motion of any kind on an e-scooter would likely be beneficial for conspicuity for detection and collision avoidance. Given the design choice between a visual motion display with biological or nonbiological motion, the recommendation would be for one consistent with biological motion. This is particularly true if the design is expected to be produced at scale, as small effects can aggregate and have a practical effect at the community level (Matz et al., 2017).

Display space can be limited, as is the case for e-scooters, reducing the ability to communicate configural form (e.g., articulated limbs), which is a major pathway for perceiving biological motion. The scooter manipulations presented in the second study focused on the local motion pathway, which has the strongest association with locomoting limbs like feet (Troje & Chang, 2023). This may partly explain why a study of retroreflective materials on pedestrians at night found that highlighting ankles of pedestrians in motion had better detection than highlighting torsos of pedestrians in motion (Balk et al., 2008). This is why the manipulation focused on apparent motion along the lower body of the PL e-scooter similar to the motion of the PL runner’s feet. However, this local motion processing pathway primarily applies to locomoting limbs, not other limbs. For example, perception of hands may rely more on the form-based mechanism of biological perception (Bromfield & Gold, 2017).

Limitations and Future Work

The primary limitation of the present study is the inability to disentangle local biological motion effects from the contribution of human body configuration to perceptual performance. Secondly, the biological motion condition presents a runner instead of a walker, which better matches the figure conditions on their typical perceived speeds as e-scooters typically move at runner speeds (or faster). Another minor limitation is that participants provided verbal report for 20/20 visual acuity (corrected allowed), so it is plausible that some participants did not have 20/20 vision. Furthermore, participants were exposed to a significant number of runners and e-scooters in the absence of a secondary task, which is not like the typical encounter rate in the real driving environment as drivers are often performing secondary tasks. If participants have increased expectations to detect figures, this may improve their detection rate and response time and obscure the true magnitude of the differences in detection accuracy between pedestrians and e-scooter riders on the road.

The biological motion advantage of the PL runner only partly replicated in Study 2, with the primary indicator being a greater degree of recognition for the runner figure in the moderately difficult distractor condition (60k). One explanation is that biological motion perception is most effective at normal walking speeds (Lange & Lappe, 2006), and the stimuli presented here all presented runners traveling at speeds consistent with an e-scooter, reducing the reliability of the biological motion effect. Another explanation may be statistical power, given an increased number of character conditions in the second study. An examination of the effect sizes of the detection accuracy (A’) for Study 1 and Study 2 between the Runner and Object (Cohen’s d = .335, 95% CI [.152, .517], Cohen’s d = .166, 95% CI [–.064, .396] indicate reduced effect sizes for Study 2, indicating that the inability to find many significant differences for Study 2 was due to both reduced effect sizes and reduced statistical power given the increased number of character conditions. Relatedly, the calculation of the false alarm rate was influenced by the increased number of signal absent trials in Study 2 (40) compared to Study 1 (24), whereas the number of signal trials per character condition remained the same for both studies (8), to have a similar ratio of total signal to nonsignal trials across studies, which may have affected A′ calculation.

Another major limitation is that these studies present relatively simple PL displays in line with classic studies investigating biological motion, instead of more realistic visual environments that may mirror a crash or near-miss events in the real world, which limits the generalizability of the findings. This approach is consistent with use-inspired basic research, in which applied problems elicit investigation in controlled basic experiments before translation to practical settings (Stokes, 1997), like basic-to-applied research programs in collision detection (DeLucia, 2013) and vigilance (Warm et al., 2008). The logical next step for future studies is to utilize more realistic stimuli or use driving simulation to investigate whether these findings with PL displays transfer to real-world settings, whether biological motion has a significant impact in a simulator for e-scooters, and whether biological motion has a greater impact for detection of e-scooter riders in environments with deprived visual information (e.g., nighttime driving, see Wood et al., 2014; 2017).

Key Points

• E-scooter riders report a relatively high prevalence of injury, and collisions with motor vehicles can be severe. Existing research on the visual perception of biological motion shows that highlighting biological motion salience increases the conspicuity of pedestrians and cyclists.

• Point light representations of a runner, e-scooter rider, and object were used to explore the relationship between the presence of biological motion and visual conspicuity.

• In the first study, participants detected and recognized the runner significantly more than the e-scooter rider or object and exhibited faster reaction times. Performance differences were reduced between the e-scooter rider and runner in the second study. Of the e-scooter conditions, the e-scooter rider displaying motion patterns inconsistent with biological motion had the worst detection performance.

• Findings were partly consistent with the hypothesis that e-scooter riders may be more difficult to visually detect by other road users due to the relative absence of local biological motion cues by riders. An alternative explanation is the higher detectability of human body configurations.

Supplemental Material

Supplemental Material - Examining Biological Motion as a Potential Factor in E-Scooter Conspicuity and Safety

Supplemental Material for Examining Biological Motion as a Potential Factor in E-Scooter Conspicuity and Safety by Marshall L. Mabry, Curtis M. Craig, Peter Easterlund, and Nichole L. Morris in Human Factors.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Funding for this study was provided by the National Science Foundation, CPS—Cyber-Physical Systems, Award Number 2038403 to Nichole L. Morris as Co-PI.

ORCID iDs

Marshall L. Mabry

Curtis M. Craig

Nichole L. Morris

Data Availability Statement

The motion capture data used in this project was obtained from ; that motion capture database was created with funding from NSF EIA-0196217.

Supplemental Material

Supplemental material for this article is available online.

Author Biographies

Marshall L. Mabry was awarded a B.A. in urban studies from the University of Minnesota, Minneapolis, Minnesota, USA, in 2022 and is currently enrolled in the urban and regional planning master’s program at the Hubert H. Humphrey School of Public Affairs. He currently works as an assistant scientist in the Human Factors Safety Laboratory in the Mechanical Engineering Department at the University of Minnesota – Twin Cities.

Curtis M. Craig was awarded a Ph.D. in experimental psychology with a human factors emphasis from Texas Tech University, Lubbock, Texas, USA, in 2016. He currently works as a research associate in the Human Factors Safety Laboratory in the Mechanical Engineering Department at the University of Minnesota – Twin Cities.

Peter Easterlund was awarded a B.S. in computer science from the University of Minnesota, Minneapolis, Minnesota, USA, in 2002. He currently works as the simulation manager for the Human Factors Safety Laboratory in the Mechanical Engineering Department at the University of Minnesota – Twin Cities.

Nichole L. Morris was awarded a Ph.D. in human factors psychology from Wichita State University, Wichita, Kansas, USA, in 2011. She is currently a research associate professor and is the director of the Human Factors Safety Laboratory in the Department of Mechanical Engineering at the University of Minnesota – Twin Cities.

References

Amoh-Gyimah

Saberi

Sarvi

(2016). Macroscopic modeling of pedestrian and bicycle crashes: A cross-comparison of estimation methods. Accident Analysis & Prevention, 93, 147–159. https://doi.org/10.1016/j.aap.2016.05.001

Austin Public Health . (2019). Epidemiology and public health preparedness division, dockless electric scooter-related injuries study. Austin Public Health. Retrieved February 26, 2024, from. https://www.austintexas.gov/sites/default/files/files/Health/Epidemiology/APH_Dockless_Electric_Scooter_Study_5-2-19.pdf

Balk

S. A.

Tyrrell

R. A.

Brooks

J. O.

Carpenter

T. L.

(2008). Highlighting human form and motion information enhances the conspicuity of pedestrians at night. Perception, 37(8), 1276–1284. https://doi.org/10.1068/p6017

Beck

L. F.

Dellinger

A. M.

O’Neil

M. E.

(2007). Motor vehicle crash injury rates by mode of travel, United States: Using exposure-based methods to quantify differences. American Journal of Epidemiology, 166(2), 212–218. https://doi.org/10.1093/aje/kwm064

Bekhit

M. N. Z.

Le Fevre

Bergin

C. J.

(2020). Regional healthcare costs and burden of injury associated with electric scooters. Injury, 51(2), 271–277. https://doi.org/10.1016/j.injury.2019.10.026

Black

A. A.

Brough

King

Bentley

L. A.

Fylan

Wood

J. M.

(2023). Making night-time pedestrians safer using innovative clothing designs. Transportation Research Part F: Traffic Psychology and Behaviour, 94(8), 321–328. https://doi.org/10.1016/j.trf.2023.03.002

Black

A. A.

Bui

Henry

Pham

Tran

Wood

J. M.

(2021). Using retro-reflective cloth to enhance drivers’ judgment of pedestrian walking direction at night-time. Journal of Safety Research, 77(8), 196–201. https://doi.org/10.1016/j.jsr.2021.03.002

Blomberg

R. D.

Hale

Preusser

D. F.

(1986). Experimental evaluation of alternative conspicuity-enhancement techniques for pedestrians and bicyclists. Journal of Safety Research, 17(1), 1–12. https://doi.org/10.1016/0022-4375(86)90002-2

Blomberg

S. N.

Rosenkrantz

O. C.

Lippert

Collatz Christensen

(2019). Injury from Electric Scooters in Copenhagen: A retrospective cohort study. BMJ Open, 9(12), Article e033988. https://doi.org/10.1136/bmjopen-2019-033988

10.

Bloom

M. B.

Noorzad

Lin

Little

Lee

E. Y.

Margulies

D. R.

Torbati

S. S.

(2021). Standing electric scooter injuries: Impact on a community. The American Journal of Surgery, 221(1), 227–232. https://doi.org/10.1016/j.amjsurg.2020.07.020

11.

Bromfield

W. D.

Gold

J. M.

(2017). Efficiencies for parts and wholes in biological-motion perception. Journal of Vision, 17(12), 21. https://doi.org/10.1167/17.12.21

12.

Brownson

A. B.

Fagan

P. V.

Dickson

Civil

I. D.

(2019). Electric scooter injuries at Auckland City Hospital. New Zealand Medical Journal, 132(1505), 62–72.

13.

Bureau of Transportation Statistics . (2024). Bikeshare (docked and dockless) and e-scooter systems by year and city served: Data & insights. Bureau of Transportation Statistics, U.S. Department of Transportation. Retrieved Feburary 23rd, 2023 from: https://data.bts.gov/Bicycles-and-Pedestrians/Bikeshare-Docked-and-Dockless-and-E-scooter-System/cqdc-cm7d/about_data

14.

Chang

D. H.

Ban

Ikegaya

Fujita

Troje

N. F.

(2018). Cortical and subcortical responses to biological motion. NeuroImage, 174, 87–96. https://doi.org/10.1016/j.neuroimage.2018.03.013

15.

Chang

D. H.

Troje

N. F.

(2009). Characterizing global and local mechanisms in biological motion perception. Journal of Vision, 9(5), 8. https://doi.org/10.1167/9.5.8

16.

Cicchino

J. B.

Kulie

P. E.

McCarthy

M. L.

(2021). Severity of e-scooter rider injuries associated with trip characteristics. Journal of Safety Research, 76(1), 256–261. https://doi.org/10.1016/j.jsr.2020.12.016

17.

Cutting

J. E.

Moore

Morrison

(1988). Masking the motions of human gait. Perception & Psychophysics, 44(4), 339–347. https://doi.org/10.3758/bf03210415

18.

DeLucia

P. R.

(2013). Effects of size on collision perception and implications for perceptual theory and transportation safety. Current Directions in Psychological Science, 22(3), 199–204. https://doi.org/10.1177/0963721412471679

19.

Dhillon

N. K.

Juillard

Barmparas

Lin

T.-L.

Kim

D. Y.

Turay

Seibold

A. R.

Kaminski

Duncan

T. K.

Diaz

Saad

Hanpeter

Benjamin

E. R.

Tillou

Demetriades

Inaba

Ley

E. J.

(2020). Electric scooter injury in Southern California trauma centers. Journal of the American College of Surgeons, 231(1), 133–138. https://doi.org/10.1016/j.jamcollsurg.2020.02.047

20.

Dittrich

W. H.

(1993). Action categories and the perception of biological motion. Perception, 22(1), 15–22. https://doi.org/10.1068/p220015

21.

Dittrich

W. H.

Troscianko

Lea

S. E.

Morgan

(1996). Perception of emotion from dynamic point-light displays represented in dance. Perception, 25(6), 727–738. https://doi.org/10.1068/p250727

22.

Edewaard

D. E.

Fekety

D. K.

Szubski

E. C.

Tyrrell

R. A.

(2020). Highlighting bicyclist biological motion enhances their conspicuity in daylight. Accident Analysis & Prevention, 142(1), Article 105575. https://doi.org/10.1016/j.aap.2020.105575

23.

Farley

K. X.

Aizpuru

Wilson

J. M.

Daly

C. A.

Xerogeanes

Gottschalk

M. B.

Wagner

E. R.

(2020). Estimated incidence of electric scooter injuries in the US from 2014 to 2019. JAMA Network Open, 3(8), Article e2014500. https://doi.org/10.1001/jamanetworkopen.2020.14500

24.

Jafari

Liu

Y. C.

(2024). Pedestrians' safety using projected time-to-collision to electric scooters. Nature Communications, 15(1), 5701. https://doi.org/10.1038/s41467-024-50049-x

25.

James

Harrois

Abback

P. S.

Moyer

J. D.

Jeantrelle

Hanouz

J. L.

Boutonnet

Geeraerts

Godier

Pottecher

Garrigue-Huet

Cotte

Pasqueron

Foucrier

Gauss

Raux

French Observatory for Major Trauma (TraumaBase) . (2023). Comparison of injuries associated with electric scooters, motorbikes, and bicycles in France, 2019-2022. JAMA Network Open, 6(6), Article e2320960. https://doi.org/10.1001/jamanetworkopen.2023.20960

26.

Janikian

G. S.

Caird

J. K.

Hagel

Reay

(2024). A scoping review of E-scooter safety: Delightful urban slalom or injury epidemic? Transportation Research Part F: Traffic Psychology and Behaviour, 101(1), 33–58. https://doi.org/10.1016/j.trf.2023.12.015

27.

Johansson

(1973). Visual perception of biological motion and a model for its analysis. Perception & Psychophysics, 14(2), 201–211. https://doi.org/10.3758/BF03212378

28.

Karpinski

Bayles

Daigle

Mantine

(2023). Comparison of motor-vehicle involved e-scooter fatalities with other traffic fatalities. Journal of Safety Research, 84(1), 61–73. https://doi.org/10.1016/j.jsr.2022.10.008

29.

Komova

Jallais

Ndiaye

Rogé

(2024). The role of innovative visibility aids on e-scooter riders’ conspicuity in daylight. Transportation Research Part F: Traffic Psychology and Behaviour, 106(1), 44–54. https://doi.org/10.1016/j.trf.2024.07.024

30.

Kozlowski

L. T.

Cutting

J. E.

(1977). Recognizing the sex of a walker from a dynamic point-light display. Perception & Psychophysics, 21(6), 575–580. https://doi.org/10.3758/bf03198740

31.

Lange

Lappe

(2006). A model of biological motion perception from configural form cues. Journal of Neuroscience: The Official Journal of the Society for Neuroscience, 26(11), 2894–2906. https://doi.org/10.1523/JNEUROSCI.4915-05.2006

32.

Leyendecker

Hackl

Leschinger

Bredow

Krane

Eysel

Müller

L. P.

Harbrecht

(2023). Lessons learned? Increasing injury severity of electric-scooter accidents over a period of one year: A monocentric follow-up study at a level 1 trauma center. European Journal of Orthopaedic Surgery and Traumatology: Orthopedie Traumatologie, 33(8), 3643–3648. https://doi.org/10.1007/s00590-023-03583-1

33.

Liew

Y. K.

Wee

C. P. J.

Pek

J. H.

(2020). New peril on our roads: A retrospective study of electric scooter-related injuries. Singapore Medical Journal, 61(2), 92–95. https://doi.org/10.11622/smedj.2019083

34.

Luoma

Schumann

Traube

E. C.

(1996). Effects of retroreflector positioning on nighttime recognition of pedestrians. Accident Analysis & Prevention, 28(3), 377–383. https://doi.org/10.1016/0001-4575(96)00004-8

35.

Mather

Radford

West

(1992). Low-level visual processing of biological motion. Proceedings. Biological Sciences, 249(1325), 149–155. https://doi.org/10.1098/rspb.1992.0097

36.

Matz

S. C.

Gladstone

J. J.

Stillwell

(2017). In a world of big data, small effects can still matter: A reply to Boyce, Daly, Hounkpatin, and Wood (2017). Psychological Science, 28(4), 547–550. https://doi.org/10.1177/0956797617697445

37.

Mehranfar

Jones

(2024). Exploring implications and current practices in e-scooter safety: A systematic review. Transportation Research Part F: Traffic Psychology and Behaviour, 107(1), 321–382. https://doi.org/10.1016/j.trf.2024.09.004

38.

Mian

Caird

J. K.

(2018). The effects of speed and orientation on recognition judgments of retro-reflectively clothed pedestrians at night. Transportation Research Part F: Traffic Psychology and Behaviour, 56(1), 185–199. https://doi.org/10.1016/j.trf.2018.04.006

39.

Neuroth

L. M.

Humphries

K. D.

Wing

J. J.

Smith

G. A.

Zhu

(2022). Motor vehicle-related electric scooter injuries in the US: A descriptive analysis of NEISS data. The American Journal of Emergency Medicine, 55(1), 1–5. https://doi.org/10.1016/j.ajem.2022.02.004

40.

Orozco-Fontalvo

Llerena

Cantillo

(2023). Dockless electric scooters: A review of a growing micromobility mode. International Journal of Sustainable Transportation, 17(4), 406–422. https://doi.org/10.1080/15568318.2022.2044097

41.

Owens

D. A.

Antonoff

R. J.

Francis

E. L.

(1994). Biological Motion and Nighttime Pedestrian Conspicuity. Human Factors, 36(4), 718–732. https://doi.org/10.1177/001872089403600411

42.

Pinto

Shiffrar

(1999). Subconfigurations of the human form in the perception of biological motion displays. Acta Psychologica, 102(2-3), 293–318. https://doi.org/10.1016/s0001-6918(99)00028-1

43.

Portland Bureau of Transportation . (2018). 2018 E-scooter findings report. City of Portland. Retrieved from. https://www.portland.gov/transportation/escooterpdx/2018-e-scooter-findings-report

44.

Rix

Demchur

N. J.

Zane

D. F.

Brown

L. H.

(2021). Injury rates per mile of travel for electric scooters versus Motor Vehicles. The American Journal of Emergency Medicine, 40(1), 166–168. https://doi.org/10.1016/j.ajem.2020.10.048

45.

Rumar

(1990). The basic driver error: late detection. Ergonomics, 33(10/11), 1281–1290. https://doi.org/10.1080/00140139008925332

46.

Sanders

R. L.

Nelson

T. A.

(2023). Results from a campus population survey of near misses, crashes, and falls while e-scooting, walking, and bicycling. Transportation Research Record: Journal of the Transportation Research Board, 2677(2), 479–489. https://doi.org/10.1177/03611981221107010

47.

Shah

N. R.

Aryal

Wen

Cherry

C. R.

(2021). Comparison of motor vehicle-involved e-scooter and bicycle crashes using standardized crash typology. Journal of Safety Research, 77(8), 217–228. https://doi.org/10.1016/j.jsr.2021.03.005

48.

Shiffrar

Lichtey

Chatterjee

S. H.

(1997). The perception of biological motion across apertures. Perception & Psychophysics, 59(1), 51–59. https://doi.org/10.3758/BF03206847

49.

Siddiqui

Abdel-Aty

Choi

(2012). Macroscopic spatial analysis of pedestrian and bicycle crashes. Accident Analysis & Prevention, 45(2), 382–391. https://doi.org/10.1016/j.aap.2011.08.003

50.

Sievert

Roen

Craig

C. M.

Morris

N. L.

(2023). A survey of electric-scooter riders’ route choice, safety perception, and helmet use. Sustainability, 15(8), 6609. https://doi.org/10.3390/su15086609

51.

Smith

K. C.

Abrams

R. A.

(2018). Motion onset really does capture attention. Attention, Perception, & Psychophysics, 80(7), 1775–1784. https://doi.org/10.3758/s13414-018-1548-1

52.

Spillmann

(2012). The current status of Gestalt rules in perceptual research: Psychophysics and neurophysiology. In Spillmann

(Ed.), On perceived motion and figural organization (pp. 191–252). Massachusetts Institute of Technology.

53.

Stanislaw

Todorov

(1999). Calculation of signal detection theory measures. Behavior Research Methods, Instruments, & Computers, 31(1), 137–149. https://doi.org/10.3758/bf03207704

54.

Stokes

D. E.

(1997). Pastuer’s quadrant: Basic science and technological innovation. Brookings Institution Press.

55.

Störmann

Klug

Nau

Verboket

R. D.

Leiblein

Müller

Schweigkofler

Hoffmann

Marzi

Lustenberger

(2020). Characteristics and injury patterns in electric-scooter related accidents—a prospective two-center report from Germany. Journal of Clinical Medicine, 9(5), 1569. https://doi.org/10.3390/jcm9051569

56.

Thurman

S. M.

(2013). Complex interactions between spatial, orientation, and motion cues for biological motion perception across visual space. Journal of Vision, 13(2), 8. https://doi.org/10.1167/13.2.8

57.

Tian

Ryan

A. D.

Craig

C. M.

Sievert

Morris

N. L.

(2022). Characteristics and risk factors for electric scooter-related crashes and injury crashes among scooter riders: A two-phase survey study. International Journal of Environmental Research and Public Health, 19(16), Article 10129. https://doi.org/10.3390/ijerph191610129

58.

Trivedi

T. K.

Liu

Antonio

A. L.

Wheaton

Kreger

Yap

Schriger

Elmore

J. G.

(2019). Injuries associated with standing electric scooter use. JAMA Network Open, 2(1), Article e187381. https://doi.org/10.1001/jamanetworkopen.2018.7381

59.

Troje

N. F.

Chang

D. H.

(2023). Life detection from biological motion. Current Directions in Psychological Science, 32(1), 26–32. https://doi.org/10.1177/09637214221128252

60.

Troje

N. F.

Westhoff

(2006). The inversion effect in biological motion perception: Evidence for a “life detector”. Current Biology: CB, 16(8), 821–824. https://doi.org/10.1016/j.cub.2006.03.022

61.

Tyrrell

R. A.

Wood

J. M.

Chaparro

Carberry

T. P.

Chu

B. S.

Marszalek

R. P.

(2009). Seeing pedestrians at night: Visual clutter does not mask biological motion. Accident Analysis & Prevention, 41(3), 506–512. https://doi.org/10.1016/j.aap.2009.02.001

62.

Warm

J. S.

Parasuraman

Matthews

(2008). Vigilance requires hard mental work and is stressful. Human Factors, 50(3), 433–441. https://doi.org/10.1518/001872008X312152

63.

Wertheimer

(2012). Investigations on gestalt principles (M. Wertheimer & K.W. Watkins, Trans.). In Spillmann

(Ed.), On perceived motion and figural organization (pp. 127–182). Massachusetts Institute of Technology.

64.

White

Guo

Han

Mollenhauer

Broaddus

Sweeney

Robinson

Novotny

Buehler

(2023). What factors contribute to e-scooter crashes: A first look using a naturalistic riding approach. Journal of Safety Research, 85(1), 182–191. https://doi.org/10.1016/j.jsr.2023.02.002

65.

Wood

J. M.

(2023). Improving the conspicuity and safety of pedestrians and cyclists on night-time roads. Clinical and Experimental Optometry, 106(3), 227–237. https://doi.org/10.1080/08164622.2023.2174001

66.

Wood

J. M.

Black

A. A.

Tyrrell

R. A.

(2022). Increasing the conspicuity of cyclists at night by using bicycle lights and clothing to highlight their biological motion to oncoming drivers. Transportation Research Part F: Traffic Psychology and Behaviour, 90(1), 326–332. https://doi.org/10.1016/j.trf.2022.09.005

67.

Wood

J. M.

Marszalek

Lacherez

Tyrrell

R. A.

(2014). Configuring retroreflective markings to enhance the night-time conspicuity of road workers. Accident Analysis & Prevention, 70(1), 209–214. https://doi.org/10.1016/j.aap.2014.03.018

68.

Wood

J. M.

Tyrrell

R. A.

Lacherez

Black

A. A.

(2017). Night-time pedestrian conspicuity: Effects of clothing on drivers’ eye movements. Ophthalmic and Physiological Optics: The Journal of the British College of Ophthalmic Opticians (Optometrists), 37(2), 184–190. https://doi.org/10.1111/opo.12351

69.

Yang

Wang

Cai

Xie

Yang

(2020). Safety of micro-mobility: Analysis of E-scooter crashes by mining news reports. Accident; Analysis and Prevention, 143(1), Article 105608. https://doi.org/10.1016/j.aap.2020.105608

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.05 MB