Pub Date : 2025-12-12DOI: 10.1016/j.aap.2025.108348
Jiateng Li, Jun Ma
Rapid advances in automated driving technology and the widespread adoption of in-vehicle information systems (IVIS) have led to an increasing prevalence of drivers engaging in non-driving-related tasks (NDRTs) during autonomous operation, thereby introducing potential safety hazards. In this study, we conducted a driving simulator experiment with 30 participants to examine the effects of IVIS NDRTs (i.e., navigation, video, audio, and reading tasks) and takeover time budgets on takeover timing, takeover quality, and visual behavior. Results from linear mixed-effects models indicate that IVIS touchscreen interactions significantly prolonged takeover time and lane change time, increased maximum lateral acceleration, and reduced minimum time-to-collision (TTC), suggesting that drivers adopted aggressive control behaviors during takeovers, which in turn elevated collision risk. Moreover, visual behavior analysis revealed an increased proportion of long glances directed away from the forward roadway and a delayed reallocation of visual attention to key regions (such as mirrors, the road, and the malfunctioning vehicle) following the takeover request. These findings enhance our understanding of human factors in automated driving and provide empirical evidence for optimizing driver-vehicle interaction protocols and improving the safety of riding in conditionally automated driving systems.
{"title":"Beyond distraction: unraveling touchscreen effects on driver takeover performance and visual attention dynamics in Level 3 automated driving","authors":"Jiateng Li, Jun Ma","doi":"10.1016/j.aap.2025.108348","DOIUrl":"10.1016/j.aap.2025.108348","url":null,"abstract":"<div><div>Rapid advances in automated driving technology and the widespread adoption of in-vehicle information systems (IVIS) have led to an increasing prevalence of drivers engaging in non-driving-related tasks (NDRTs) during autonomous operation, thereby introducing potential safety hazards. In this study, we conducted a driving simulator experiment with 30 participants to examine the effects of IVIS NDRTs (i.e., navigation, video, audio, and reading tasks) and takeover time budgets on takeover timing, takeover quality, and visual behavior. Results from linear mixed-effects models indicate that IVIS touchscreen interactions significantly prolonged takeover time and lane change time, increased maximum lateral acceleration, and reduced minimum time-to-collision (TTC), suggesting that drivers adopted aggressive control behaviors during takeovers, which in turn elevated collision risk. Moreover, visual behavior analysis revealed an increased proportion of long glances directed away from the forward roadway and a delayed reallocation of visual attention to key regions (such as mirrors, the road, and the malfunctioning vehicle) following the takeover request. These findings enhance our understanding of human factors in automated driving and provide empirical evidence for optimizing driver-vehicle interaction protocols and improving the safety of riding in conditionally automated driving systems.</div></div>","PeriodicalId":6926,"journal":{"name":"Accident; analysis and prevention","volume":"226 ","pages":"Article 108348"},"PeriodicalIF":6.2,"publicationDate":"2025-12-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145735405","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-11DOI: 10.1016/j.aap.2025.108347
Peipei Guo , Fangtong Jiao , Zhigang Du , Feng Sun , Xinke Tang
The multi-entry underpass road tunnels is affected by various factors, including long downhill approaches outside the tunnel, monotonous visual environments inside the tunnel, underground merging of the main and secondary roads, and limited sight distance and sight zone. These combined conditions can lead to perception and judgment errors among drivers, significantly increasing the accident risk of rear-end and lateral crashes. This study used video data from a real vehicle test and conducted a subjective perception experiment with a driving simulator. It collected key indicators related to crash accident risk and prevention, including Identify Merging Time (IMT), Perceive Hazard Time (PHT), and Assess Safety Time (AST), to analyze the dynamic perception of risk and safety at the entrances of the main and secondary roads under 6 different speeds. And a Linear Mixed Model (LMM) was applied to evaluate the effect of speed on each indicator. Results showed that IMT decreased with increasing speed for both main and secondary roads, with the main road exhibited the highest Identify Merging Delay Rate (IMDR) at 38.667 %, indicating that drivers traveling at higher speeds struggled to identify the underground merging area in time. The Perceive Hazard Distance (PHD) for both main and secondary roads extended with increasing speed. Compared to the main road, drivers on the secondary road perceived hazards earlier within 38.167 to 46.683 m downstream of the physical gore point. This earlier perception was related to their frequent use of rearview mirrors to assess merging opportunities and the expanded sight zone in the secondary road merging area. Through LMM analysis, secondary road drivers’ PHD is less dependent on speed and is more influenced by the merging process itself. Overall, at higher speeds, reaction time is notably reduced, leading to delayed identification, hazard perception, and safety assessment. Hence, these findings provide valuable references for optimizing underground merging area design and enhancing drivers’ safety perception in multi-entry underpass road tunnels.
{"title":"Drivers’ dynamic perception of accident risk and safety in underground road merging areas","authors":"Peipei Guo , Fangtong Jiao , Zhigang Du , Feng Sun , Xinke Tang","doi":"10.1016/j.aap.2025.108347","DOIUrl":"10.1016/j.aap.2025.108347","url":null,"abstract":"<div><div>The multi-entry underpass road tunnels is affected by various factors, including long downhill approaches outside the tunnel, monotonous visual environments inside the tunnel, underground merging of the main and secondary roads, and limited sight distance and sight zone. These combined conditions can lead to perception and judgment errors among drivers, significantly increasing the accident risk of rear-end and lateral crashes. This study used video data from a real vehicle test and conducted a subjective perception experiment with a driving simulator. It collected key indicators related to crash accident risk and prevention, including Identify Merging Time (IMT), Perceive Hazard Time (PHT), and Assess Safety Time (AST), to analyze the dynamic perception of risk and safety at the entrances of the main and secondary roads under 6 different speeds. And a Linear Mixed Model (LMM) was applied to evaluate the effect of speed on each indicator. Results showed that IMT decreased with increasing speed for both main and secondary roads, with the main road exhibited the highest Identify Merging Delay Rate (IMDR) at 38.667 %, indicating that drivers traveling at higher speeds struggled to identify the underground merging area in time. The Perceive Hazard Distance (PHD) for both main and secondary roads extended with increasing speed. Compared to the main road, drivers on the secondary road perceived hazards earlier within 38.167 to 46.683 m downstream of the physical gore point. This earlier perception was related to their frequent use of rearview mirrors to assess merging opportunities and the expanded sight zone in the secondary road merging area. Through LMM analysis, secondary road drivers’ PHD is less dependent on speed and is more influenced by the merging process itself. Overall, at higher speeds, reaction time is notably reduced, leading to delayed identification, hazard perception, and safety assessment. Hence, these findings provide valuable references for optimizing underground merging area design and enhancing drivers’ safety perception in multi-entry underpass road tunnels.</div></div>","PeriodicalId":6926,"journal":{"name":"Accident; analysis and prevention","volume":"226 ","pages":"Article 108347"},"PeriodicalIF":6.2,"publicationDate":"2025-12-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145735351","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-11DOI: 10.1016/j.aap.2025.108334
Qingwen Pu , Kun Xie , Hongyu Guo , Yuan Zhu
Understanding vehicle–pedestrian interactions at urban intersections is critical for enhancing traffic safety. This study aims to model the interactive crash avoidance behavior of road users in near-miss scenarios, addressing the complexities of their decision-making process. Utilizing high-resolution trajectory data collected by unmanned aerial vehicles (UAV), this study proposed a multi-agent state-space Transformer enhanced deep deterministic policy gradient (MA-SST-DDPG) framework to model the vehicle–pedestrian interactions in near-miss scenarios. The framework integrates the state-space model for capturing long-term temporal dependencies and Transformers for dynamically prioritizing critical features, enhancing its ability to learn from rare safety–critical scenarios. The MA-SST-DDPG framework effectively learned sequential decision-making over continuous action spaces, effectively prioritizing critical states and capturing dynamic interactions in vehicle–pedestrian near-miss scenarios. Compared to alternative approaches, it demonstrated superior performance in simulating realistic evasive behaviors. Cross-dataset evaluation confirmed the generalizability of the proposed model on external datasets. Additionally, we employed the proposed model to generate vehicle–pedestrian interactions under varying combinations of initial speeds. Results showed that the simulated interactions successfully replicated the dynamics of real-world near-miss events. Higher initial vehicle and pedestrian speeds were linked to increased conflict rates. Moreover, pedestrians were more likely to yield when vehicles traveled faster and pedestrians walked slower, whereas slower vehicles tended to yield to faster-moving pedestrians. The outcomes of this study can facilitate the development of safety-aware simulations that closely mimic interactive crash avoidance behaviors of road users, paving the way for exploring proactive measures to prevent crashes.
{"title":"Modeling interactive crash avoidance behaviors: A multi-agent state-space transformer-enhanced reinforcement learning framework","authors":"Qingwen Pu , Kun Xie , Hongyu Guo , Yuan Zhu","doi":"10.1016/j.aap.2025.108334","DOIUrl":"10.1016/j.aap.2025.108334","url":null,"abstract":"<div><div>Understanding vehicle–pedestrian interactions at urban intersections is critical for enhancing traffic safety. This study aims to model the interactive crash avoidance behavior of road users in near-miss scenarios, addressing the complexities of their decision-making process. Utilizing high-resolution trajectory data collected by unmanned aerial vehicles (UAV), this study proposed a multi-agent state-space Transformer enhanced deep deterministic policy gradient (MA-SST-DDPG) framework to model the vehicle–pedestrian interactions in near-miss scenarios. The framework integrates the state-space model for capturing long-term temporal dependencies and Transformers for dynamically prioritizing critical features, enhancing its ability to learn from rare safety–critical scenarios. The MA-SST-DDPG framework effectively learned sequential decision-making over continuous action spaces, effectively prioritizing critical states and capturing dynamic interactions in vehicle–pedestrian near-miss scenarios. Compared to alternative approaches, it demonstrated superior performance in simulating realistic evasive behaviors. Cross-dataset evaluation confirmed the generalizability of the proposed model on external datasets. Additionally, we employed the proposed model to generate vehicle–pedestrian interactions under varying combinations of initial speeds. Results showed that the simulated interactions successfully replicated the dynamics of real-world near-miss events. Higher initial vehicle and pedestrian speeds were linked to increased conflict rates. Moreover, pedestrians were more likely to yield when vehicles traveled faster and pedestrians walked slower, whereas slower vehicles tended to yield to faster-moving pedestrians. The outcomes of this study can facilitate the development of safety-aware simulations that closely mimic interactive crash avoidance behaviors of road users, paving the way for exploring proactive measures to prevent crashes.</div></div>","PeriodicalId":6926,"journal":{"name":"Accident; analysis and prevention","volume":"226 ","pages":"Article 108334"},"PeriodicalIF":6.2,"publicationDate":"2025-12-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145735318","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-08DOI: 10.1016/j.aap.2025.108338
Kaiming Guan, Junyi Zhang, Wei Ye, Ying Jiang
Existing traffic conflict models face challenges in handling minority class samples and capturing dynamic interactions in complex traffic scenarios. These limitations hinder model generalization and real-world applicability. This study employs an enhanced Two-dimensional Time-to-collision (2D-TTC) metric combined with vehicle interaction relationships to predict traffic conflicts of multiple patterns. To address imbalance in conflict and non-conflict events, both undersampling and oversampling techniques are employed, while a generative adversarial network with self-attention layers is leveraged to overcome the shortcomings of oversampling methods. Indeed, this approach proved highly effective, elevating the model’s F1-score from 76.35 % with undersampling alone to 94.21 %. Additionally, several machine learning and deep learning models are compared, with the hypergraph attention network combined with Shapley additive explanations (S-HGAT) demonstrating the strongest learning capability. Furthermore, vehicle speed is identified as the most influential factor associated with traffic conflicts. A comprehensive re-evaluation of feature combinations reveals that the top six features—vehicle speed, the number of vehicles ahead, the standard deviation and the average of vehicle speeds within the traffic flow, distance with the road markings, and peak traffic hour indicators—result in the highest model F1-score of 98.41 % and accuracy of 97.66 %. Finally, the real-world implications of these findings are discussed.
{"title":"Solution to data imbalance and complex interactions in traffic conflict modeling: a hypergraph and generative AI approach","authors":"Kaiming Guan, Junyi Zhang, Wei Ye, Ying Jiang","doi":"10.1016/j.aap.2025.108338","DOIUrl":"10.1016/j.aap.2025.108338","url":null,"abstract":"<div><div>Existing traffic conflict models face challenges in handling minority class samples and capturing dynamic interactions in complex traffic scenarios. These limitations hinder model generalization and real-world applicability. This study employs an enhanced Two-dimensional Time-to-collision (2D-TTC) metric combined with vehicle interaction relationships to predict traffic conflicts of multiple patterns. To address imbalance in conflict and non-conflict events, both undersampling and oversampling techniques are employed, while a generative adversarial network with self-attention layers is leveraged to overcome the shortcomings of oversampling methods. Indeed, this approach proved highly effective, elevating the model’s F1-score from 76.35 % with undersampling alone to 94.21 %. Additionally, several machine learning and deep learning models are compared, with the hypergraph attention network combined with Shapley additive explanations (S-HGAT) demonstrating the strongest learning capability. Furthermore, vehicle speed is identified as the most influential factor associated with traffic conflicts. A comprehensive re-evaluation of feature combinations reveals that the top six features—vehicle speed, the number of vehicles ahead, the standard deviation and the average of vehicle speeds within the traffic flow, distance with the road markings, and peak traffic hour indicators—result in the highest model F1-score of 98.41 % and accuracy of 97.66 %. Finally, the real-world implications of these findings are discussed.</div></div>","PeriodicalId":6926,"journal":{"name":"Accident; analysis and prevention","volume":"226 ","pages":"Article 108338"},"PeriodicalIF":6.2,"publicationDate":"2025-12-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145713034","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-05DOI: 10.1016/j.aap.2025.108339
Zhigang Wu , Meng Li , Yanyong Guo , Zhibin Li , Shunchao Wang
Large-scale traffic accidents are often triggered by sudden shockwaves in congested flow, typically caused by unpredictable driving behaviors. The collaboration among connected and autonomous vehicles (CAVs) offer potential to mitigating traffic congestion and accidents, yet it remains vulnerable to failures in vehicle behavior coordination due to unstable long-range communication. To address these issues, this study proposes a Cross-Network Collaboration-based Congestion Mitigation (CNC-CM) framework, which establishes a feedback response mechanism between the traffic system and the communication network. At the communication layer, a distance-to-delay interval backtracking algorithm is developed to optimize long-range hybrid communication routing, ensuring timely and reliable command delivery under varying network conditions. At the traffic control layer, a multi-scale cooperative strategy is designed: a micro-level barrier consensus control restrains disruptive lane-changing by human-driven vehicles (HDVs), while a macro-level delay-corrected cruising control eliminates stop-and-go waves within enclosed congestion clusters. By integrating communication constraints into traffic control decisions, this cross-scale, multi-layer approach proactively dissipates incipient traffic jams before they escalate into safety hazards. Simulation results demonstrate that the proposed control framework enhances driving safety by over 54.11% through completely eliminating traffic congestion, while also significantly improving traffic efficiency, reducing energy consumption, and enhancing communication quality. Notably, the framework maintains robust performance even under low CAV penetration rates, confirming its effectiveness in mixed traffic environments with unpredictable human driving behaviors.
{"title":"A cross-scale traffic-communication control framework for improving safety through proactive congestion mitigation in mixed traffic","authors":"Zhigang Wu , Meng Li , Yanyong Guo , Zhibin Li , Shunchao Wang","doi":"10.1016/j.aap.2025.108339","DOIUrl":"10.1016/j.aap.2025.108339","url":null,"abstract":"<div><div>Large-scale traffic accidents are often triggered by sudden shockwaves in congested flow, typically caused by unpredictable driving behaviors. The collaboration among connected and autonomous vehicles (CAVs) offer potential to mitigating traffic congestion and accidents, yet it remains vulnerable to failures in vehicle behavior coordination due to unstable long-range communication. To address these issues, this study proposes a Cross-Network Collaboration-based Congestion Mitigation (CNC-CM) framework, which establishes a feedback response mechanism between the traffic system and the communication network. At the communication layer, a distance-to-delay interval backtracking algorithm is developed to optimize long-range hybrid communication routing, ensuring timely and reliable command delivery under varying network conditions. At the traffic control layer, a multi-scale cooperative strategy is designed: a micro-level barrier consensus control restrains disruptive lane-changing by human-driven vehicles (HDVs), while a macro-level delay-corrected cruising control eliminates stop-and-go waves within enclosed congestion clusters. By integrating communication constraints into traffic control decisions, this cross-scale, multi-layer approach proactively dissipates incipient traffic jams before they escalate into safety hazards. Simulation results demonstrate that the proposed control framework enhances driving safety by over 54.11% through completely eliminating traffic congestion, while also significantly improving traffic efficiency, reducing energy consumption, and enhancing communication quality. Notably, the framework maintains robust performance even under low CAV penetration rates, confirming its effectiveness in mixed traffic environments with unpredictable human driving behaviors.</div></div>","PeriodicalId":6926,"journal":{"name":"Accident; analysis and prevention","volume":"226 ","pages":"Article 108339"},"PeriodicalIF":6.2,"publicationDate":"2025-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145665619","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-05DOI: 10.1016/j.aap.2025.108336
Seungyeon Lee , Eun Hak Lee
Colored lane markings are a recent traffic safety intervention in South Korea, designed to improve driver awareness and visual guidance. This study aims to evaluate their effectiveness in reducing traffic crashes. Specifically, 82 road segments in Seoul where the markings were installed were analyzed by comparing crash trends before and after the intervention using data from 2010 to 2024. To estimate the intervention’s effect, a counterfactual analysis was conducted by constructing a baseline scenario representing crash trends in the absence of the intervention. The causal impact of the colored markings was then identified by comparing this baseline with observed outcomes. The results show that the implementation of colored lane markings led to an average 26.7 % reduction in crash rates at statistically significant sites. To identify where the intervention was most effective, the relationship between surrounding land use and observed safety outcomes was examined. The analysis indicates that the markings were more effective on highways and arterial roads, which tend to have higher speeds and simpler traffic conditions. In contrast, roads in dense urban areas showed limited improvements. This outcome is attributable to complex traffic conditions and high levels of visual and environmental clutter. Taken together, these findings suggest that the intervention is highly effective and provides safety benefits on arterial networks.
{"title":"Do colored lane markings improve road safety? Causal evidence from Seoul","authors":"Seungyeon Lee , Eun Hak Lee","doi":"10.1016/j.aap.2025.108336","DOIUrl":"10.1016/j.aap.2025.108336","url":null,"abstract":"<div><div>Colored lane markings are a recent traffic safety intervention in South Korea, designed to improve driver awareness and visual guidance. This study aims to evaluate their effectiveness in reducing traffic crashes. Specifically, 82 road segments in Seoul where the markings were installed were analyzed by comparing crash trends before and after the intervention using data from 2010 to 2024. To estimate the intervention’s effect, a counterfactual analysis was conducted by constructing a baseline scenario representing crash trends in the absence of the intervention. The causal impact of the colored markings was then identified by comparing this baseline with observed outcomes. The results show that the implementation of colored lane markings led to an average 26.7 % reduction in crash rates at statistically significant sites. To identify where the intervention was most effective, the relationship between surrounding land use and observed safety outcomes was examined. The analysis indicates that the markings were more effective on highways and arterial roads, which tend to have higher speeds and simpler traffic conditions. In contrast, roads in dense urban areas showed limited improvements. This outcome is attributable to complex traffic conditions and high levels of visual and environmental clutter. Taken together, these findings suggest that the intervention is highly effective and provides safety benefits on arterial networks.</div></div>","PeriodicalId":6926,"journal":{"name":"Accident; analysis and prevention","volume":"226 ","pages":"Article 108336"},"PeriodicalIF":6.2,"publicationDate":"2025-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145684091","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-03DOI: 10.1016/j.aap.2025.108335
John N. Ivan , Yaohua Zhang , Nalini Ravishanker
The SHRP2 Naturalistic Driving Study (NDS) data were used to investigate association between actual driving speeds before known crashes and at other times. Associations were evaluated for the same driver at a location where a crash occurred and similar locations where crashes did not occur, relative to the speeds of other drivers at those locations. It was found that an increase in the speed differential relative to other drivers at the same location between 6 and 10 s before a crash occurred was significantly associated with a crash occurring. The quantile of the average speed over that five-second period served as a better predictor than the quantile of the maximum speed. Crashes were also more associated with road locations classified as limited access highways, minor arterials, and major collectors. These findings are consistent across different drivers and types of road locations. The best-performing model classified all of the crashes in the dataset perfectly, and less than half of the cases classified as crashes were not crashes. This suggests an ability to identify conditions that are at least 50 percent likely to result in a crash. The results could be used by road agencies to identify observed vehicle speed variations that are likely to result in crashes, as well as by vehicle manufacturers to develop algorithms for identifying high-risk conditions for crashes considering speeds of other vehicles in the vicinity.
{"title":"Evaluation of association between observed driving speeds and the occurrence of crashes using naturalistic driving study data","authors":"John N. Ivan , Yaohua Zhang , Nalini Ravishanker","doi":"10.1016/j.aap.2025.108335","DOIUrl":"10.1016/j.aap.2025.108335","url":null,"abstract":"<div><div>The SHRP2 Naturalistic Driving Study (NDS) data were used to investigate association between actual driving speeds before known crashes and at other times. Associations were evaluated for the same driver at a location where a crash occurred and similar locations where crashes did not occur, relative to the speeds of other drivers at those locations. It was found that an increase in the speed differential relative to other drivers at the same location between 6 and 10 s before a crash occurred was significantly associated with a crash occurring. The quantile of the average speed over that five-second period served as a better predictor than the quantile of the maximum speed. Crashes were also more associated with road locations classified as limited access highways, minor arterials, and major collectors. These findings are consistent across different drivers and types of road locations. The best-performing model classified all of the crashes in the dataset perfectly, and less than half of the cases classified as crashes were not crashes. This suggests an ability to identify conditions that are at least 50 percent likely to result in a crash. The results could be used by road agencies to identify observed vehicle speed variations that are likely to result in crashes, as well as by vehicle manufacturers to develop algorithms for identifying high-risk conditions for crashes considering speeds of other vehicles in the vicinity.</div></div>","PeriodicalId":6926,"journal":{"name":"Accident; analysis and prevention","volume":"225 ","pages":"Article 108335"},"PeriodicalIF":6.2,"publicationDate":"2025-12-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145676087","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-03DOI: 10.1016/j.aap.2025.108337
Jianlin Li , Jun Zhang , Shuchao Cao , Xiangxia Ren , Weiguo Song , Eric Wai Ming Lee
Pedestrian following behavior has a significant impact on crowd dynamics within transportation hubs, where dense and heterogeneous passenger flows pose substantial traffic risk challenges. However, empirical research on this behavior in such complex environments remains scarce, and many existing models still rely on subjective assumptions. This study bridges this gap through controlled experiments to investigate pedestrian following behavior during walking and running, which are typical movement states in transportation hub scenarios. The results demonstrate that pedestrians exhibit a lower willingness to follow others when running compared to walking. After ceasing one following behavior, more than 70% of pedestrians will initiate the next following within 0.5 s. Walking pedestrians tend to follow the individual within 2.49 m ahead with an angle rang of −53.77° to 50.25°, while the running pedestrians prefer to follow the one within 1.99 m ahead with an angle range of −80.83° to 57.21°, parameters that can inform spatial risk assessment in hub functional zones. Notably, no clear evidence shows that followers prefer pedestrians with larger speed differences or those whose movement directions align closely with their desired velocity direction. Additionally, during the following process, followers may switch their targets or cease following them. The study finds that “the distance between the follower and the leader” and “the number of pedestrians within the rectangular area formed by the positions of the follower and the leader” are the main driving factors for changes in the following behavior of followers in both walking and running states. Specifically, as these two factors increase, the probability of followers changing their following behavior also rises, which is vital for developing safety control strategies during passenger transfers. These findings are further compared with the assumptions about following behavior in previous models. This study enhances the understanding of pedestrian dynamics and aims to facilitate the integration of following behavior into crowd dynamics models, thereby improving the accuracy of evacuation models and supporting traffic risk prevention in hub systems.
{"title":"Experimental study on the following behavior of pedestrians encountering those who go against the flow","authors":"Jianlin Li , Jun Zhang , Shuchao Cao , Xiangxia Ren , Weiguo Song , Eric Wai Ming Lee","doi":"10.1016/j.aap.2025.108337","DOIUrl":"10.1016/j.aap.2025.108337","url":null,"abstract":"<div><div>Pedestrian following behavior has a significant impact on crowd dynamics within transportation hubs, where dense and heterogeneous passenger flows pose substantial traffic risk challenges. However, empirical research on this behavior in such complex environments remains scarce, and many existing models still rely on subjective assumptions. This study bridges this gap through controlled experiments to investigate pedestrian following behavior during walking and running, which are typical movement states in transportation hub scenarios.<!--> <!-->The<!--> <!-->results demonstrate that pedestrians exhibit<!--> <!-->a<!--> <!-->lower willingness to follow others when running compared to walking. After ceasing one following behavior, more than 70% of pedestrians will initiate the next following within 0.5 s. Walking pedestrians tend to follow the individual within 2.49 m ahead with an angle rang of −53.77° to 50.25°, while the running pedestrians prefer to follow the one within 1.99 m ahead with an angle range of −80.83° to 57.21°, parameters that can inform spatial risk assessment in hub functional zones. Notably, no clear evidence shows that followers prefer pedestrians with larger speed differences or those whose movement directions align closely with their desired velocity direction. Additionally, during the following process, followers may switch their targets or cease following them. The study finds that “the distance between the follower and the leader” and “the number of pedestrians within the rectangular area formed by the positions of the follower and the leader” are the main driving factors for changes in the following behavior of followers in both walking and running states. Specifically, as these two factors increase, the probability of followers changing their following behavior also rises, which is vital for developing safety control strategies during passenger transfers. These findings are further compared with the assumptions about following behavior in previous models. This study enhances the understanding of pedestrian dynamics and aims to facilitate the integration of following behavior into crowd dynamics models, thereby improving the accuracy of evacuation models and supporting traffic risk prevention in hub systems.</div></div>","PeriodicalId":6926,"journal":{"name":"Accident; analysis and prevention","volume":"225 ","pages":"Article 108337"},"PeriodicalIF":6.2,"publicationDate":"2025-12-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145676145","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-03DOI: 10.1016/j.aap.2025.108332
Benjamin P. Krbavac , Rory England , Séan Mitchell , Paul Sherratt , Kevin Gildea , Jon Farmer
Mild traumatic brain injury (mTBI) is a frequent but underreported consequence of professional cycling crashes, yet current helmet testing standards primarily simulate head-first impacts, and their representation of real-world head impact scenarios is unclear. This study explores crash typology of professional cycling crashes involving head-ground contact through systematic video analysis of 128 head impacts occurring between 2012 and 2024. Most head impacts occurred during road races (113/128, 88 %) and were associated with multi-cyclist collisions rather than single-cyclist crashes, with topple-over crashes representing the most common mechanism (49 %), followed by skid-outs. Riders predominantly landed front or front-side relative to their direction of travel, with 66 % of impacts occurring in a sideways body posture, and head contact most frequently involved the helmet’s side and rim regions (>50 % of impacts). Notably, body-first head impacts dominated the crash profiles (92 %), with the torso or arms contacting the ground before the head, while direct head-first impacts comprised 8 % of cases. Impact severity was distributed relatively evenly across low (30 %), medium (33 %), and high (36 %) categories, with collision-related crashes being more likely to result in high-severity outcomes than non-contact crashes. These findings reveal a potential mismatch between current helmet testing protocols and the predominant mechanisms observed in professional cycling crashes. Video-based analysis provides critical insights into impact mechanisms that are overlooked by traditional injury reporting methods, particularly highlighting the prevalence of body-first impacts and side-rim head impacts. This crash typology may provide a foundation for future biomechanical studies and could support the development of helmet testing methods that better represent real-world cycling impact scenarios.
{"title":"Crash typology of professional cycling crashes","authors":"Benjamin P. Krbavac , Rory England , Séan Mitchell , Paul Sherratt , Kevin Gildea , Jon Farmer","doi":"10.1016/j.aap.2025.108332","DOIUrl":"10.1016/j.aap.2025.108332","url":null,"abstract":"<div><div>Mild traumatic brain injury (mTBI) is a frequent but underreported consequence of professional cycling crashes, yet current helmet testing standards primarily simulate head-first impacts, and their representation of real-world head impact scenarios is unclear. This study explores crash typology of professional cycling crashes involving head-ground contact through systematic video analysis of 128 head impacts occurring between 2012 and 2024. Most head impacts occurred during road races (113/128, 88 %) and were associated with multi-cyclist collisions rather than single-cyclist crashes, with topple-over crashes representing the most common mechanism (49 %), followed by skid-outs. Riders predominantly landed front or front-side relative to their direction of travel, with 66 % of impacts occurring in a sideways body posture, and head contact most frequently involved the helmet’s side and rim regions (>50 % of impacts). Notably, body-first head impacts dominated the crash profiles (92 %), with the torso or arms contacting the ground before the head, while direct head-first impacts comprised 8 % of cases. Impact severity was distributed relatively evenly across low (30 %), medium (33 %), and high (36 %) categories, with collision-related crashes being more likely to result in high-severity outcomes than non-contact crashes. These findings reveal a potential mismatch between current helmet testing protocols and the predominant mechanisms observed in professional cycling crashes. Video-based analysis provides critical insights into impact mechanisms that are overlooked by traditional injury reporting methods, particularly highlighting the prevalence of body-first impacts and side-rim head impacts. This crash typology may provide a foundation for future biomechanical studies and could support the development of helmet testing methods that better represent real-world cycling impact scenarios.</div></div>","PeriodicalId":6926,"journal":{"name":"Accident; analysis and prevention","volume":"225 ","pages":"Article 108332"},"PeriodicalIF":6.2,"publicationDate":"2025-12-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145676103","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-03DOI: 10.1016/j.aap.2025.108328
Xin Gao , Quan Li , Siyuan Liu , Yiran Luo , Bo Zhang , Wei Lu , Zheng Wang , Qing Zhou , Bingbing Nie
Pedestrians, as vulnerable road users, face significant safety risks in traffic environments. In hazardous traffic scenarios, they often exhibit collision-avoidance behaviors that significantly influence vehicle interactions and potential injury risks. However, the inherent uncertainty and complexity of influencing factors make accurate modeling of pedestrian behavior in safety–critical scenarios particularly challenging. To address this, we propose a comprehensive model for pedestrian dynamic behaviors in such scenarios, which is subsequently implemented in a virtual vehicle safety testing platform to enable realistic pedestrian-vehicle interactions. The developed model systematically captures the process of pedestrian risk perception, avoidance decision-making, and kinematic response, ensuring high behavioral fidelity by leveraging real pedestrian data collected in a virtual reality experiment. Utilizing actual accident data, the constructed platform is capable of generating imminent yet realistic pedestrian-vehicle pre-crash scenarios. Furthermore, it integrates pedestrian injury risk into a unified safety evaluation framework that synergizes active and passive safety assessments. Validation with real-world data demonstrates the effectiveness of the proposed pedestrian model and testing platform. Across 4,200 simulated pedestrian-vehicle interactions, pedestrians’ active avoidance of oncoming vehicles reduced collisions by 27.77 % relative to a no-avoidance baseline, highlighting the necessity of incorporating dynamic behavior modeling into virtual testing. The virtual testing framework proposed herein enables practical implementation of the high-fidelity pedestrian model, offering a pathway for autonomous vehicles to better interpret pedestrian behaviors and enhance interactive safety.
{"title":"Pedestrian modeling with realistic dynamic behaviors and its application in virtual safety testing for autonomous vehicles","authors":"Xin Gao , Quan Li , Siyuan Liu , Yiran Luo , Bo Zhang , Wei Lu , Zheng Wang , Qing Zhou , Bingbing Nie","doi":"10.1016/j.aap.2025.108328","DOIUrl":"10.1016/j.aap.2025.108328","url":null,"abstract":"<div><div>Pedestrians, as vulnerable road users, face significant safety risks in traffic environments. In hazardous traffic scenarios, they often exhibit collision-avoidance behaviors that significantly influence vehicle interactions and potential injury risks. However, the inherent uncertainty and complexity of influencing factors make accurate modeling of pedestrian behavior in safety–critical scenarios particularly challenging. To address this, we propose a comprehensive model for pedestrian dynamic behaviors in such scenarios, which is subsequently implemented in a virtual vehicle safety testing platform to enable realistic pedestrian-vehicle interactions. The developed model systematically captures the process of pedestrian risk perception, avoidance decision-making, and kinematic response, ensuring high behavioral fidelity by leveraging real pedestrian data collected in a virtual reality experiment. Utilizing actual accident data, the constructed platform is capable of generating imminent yet realistic pedestrian-vehicle pre-crash scenarios. Furthermore, it integrates pedestrian injury risk into a unified safety evaluation framework that synergizes active and passive safety assessments. Validation with real-world data demonstrates the effectiveness of the proposed pedestrian model and testing platform. Across 4,200 simulated pedestrian-vehicle interactions, pedestrians’ active avoidance of oncoming vehicles reduced collisions by 27.77 % relative to a no-avoidance baseline, highlighting the necessity of incorporating dynamic behavior modeling into virtual testing. The virtual testing framework proposed herein enables practical implementation of the high-fidelity pedestrian model, offering a pathway for autonomous vehicles to better interpret pedestrian behaviors and enhance interactive safety.</div></div>","PeriodicalId":6926,"journal":{"name":"Accident; analysis and prevention","volume":"225 ","pages":"Article 108328"},"PeriodicalIF":6.2,"publicationDate":"2025-12-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145676090","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号