2021

Byeongjoon Noh, Korea Advanced Institute of Science and Technology (KAIST)
Dongho Ka, Korea Advanced Institute of Science and Technology (KAIST)
David Lee, Korea Advanced Institute of Science and Technology (KAIST)
Hwasoo Yeo (hwasoo@gmail.com), Korea Advanced Institute of Science and Technology (KAIST)

Even though advancement of information communication technologies (ICT) leads to improving the quality of life in the overall fields, road traffic accidents globally posed a severe threat to human lives and have become a leading cause of premature deaths. In particular, the casualties of pedestrian crossing the road present a major cause in vehicle-pedestrian accidents in South Korea, but we lack dense behavioral data to understand the risk they face. In this paper, we propose a new analytical system for potential pedestrian risk event based on video footage obtained by road security cameras already deployed at unsignalized crosswalks. Our system can automatically extract the behavioral features of vehicles and pedestrians, affecting the likelihood of the potentially dangerous situations after detecting them into individual objects. In our experiments, we handle only four behavioral features such as vehicle velocity, pedestrian position, vehicle-pedestrian distance, and vehicle-crosswalk distance. Then, in order to show how they can be useful for monitoring the traffic behaviors on the road, we visualize and interpret them to show how they may or may not contribute to potential pedestrian risks of these crosswalks: 1) by analyzing car velocity changes near the crosswalk when there are no pedestrians present; and 2) analyzing car velocities by vehicle-pedestrian distances when pedestrians are on the crosswalk. We validate the feasibility of the propose system by applying our system to multiple unsignalized crosswalks in Osan city, South Korea.

Maged Gouda (mgouda@ualberta.ca), University of Alberta
Jie Fan, University of Alberta
Shewkar Ibrahim, City of Edmonton
Karim El-Basyouny, University of Alberta

Physical distancing is proven to be the most effective countermeasure to slow the spread of COVID-19 and is necessary considering there is currently no cure or vaccine available. The concept of redesigning public spaces to encourage physical distancing is being tested around the world. All over Canada, municipalities are reallocating underutilized road lanes for active modes of transportation, such as walking and cycling. We evaluated the usage and benefit of these shared spaces to ensure redesign efforts are optimally allocated. We analyzed CCTV footage before and after the change, from April 7-13, of two locations in the City of Edmonton using automated computer vision techniques. We detected and recorded physical distancing violations and generated trajectory maps of all road users. A manual review of the footage was performed to verify the automated analysis. From trajectory analysis, it was found that the redesign was utilized effectively by road-users and improved physical distancing compliance. The proposed framework provides an innovative tool to automatically gather, extract, share, and analyze real-world data to improve response to the COVID-19 pandemic as well as future outbreaks of contagious disease.

Arunabha Banerjee, Indian Institute of Technology, Guwahati
Anuj Budhkar, Indian Institute of Engineering Science and Technology
Akhilesh Maurya, Indian Institute of Technology, Guwahati

The present study developed a novel method of semi-manual data extraction technique for pedestrian position (subsequently speed and trajectory) data using front inclined camera angle as many times top-down angle is unavailable due to height restrictions. Vanishing point method was used to extract data for pedestrian body dimensions across vertical and pedestrian-specific trajectory planes. Using the trajectory and body dimension information, the spacings maintained between pedestrians were estimated. Subsequently, JUPedSim software was used for extracting the fundamental macroscopic properties (speed, flow density) from the pedestrian position data, and the results were compared using classical and voronoi approaches. The developed technique was able to extract the pedestrian trajectories with an error (standard deviation) of 0.039 m at the exit section and 0.11 m at the entry section. The average lateral, longitudinal and diagonal clearances maintained by a pedestrian to walk comfortably were observed as 0.25m, 0.90m and 1.07m respectively. The comparison study for voronoi vs classical approach showed that voronoi approach had lower fluctuations in estimation of macroscopic parameters. This semi-manual technique could be helpful in study of pedestrian gap maintaining behavior and establishment of different threshold levels for the crowd management. Such approach can directly be applied to CCTV footage (which mostly gives front inclined views of camera) to ensure the social distancing in the current ongoing pandemic COVID-19. Further, the proposed approach can be made automated using the image processing tools for different applications like determination of pedestrian stream characteristics, level of services, safety evaluation and ensuring social distancing.

Frank Appiah, Portland State University
Sirisha Kothuri (skothuri@pdx.edu), Portland State University
Christopher Monsere, Portland State University

RRFBs have proven to be a useful tool for improving driver yielding and safety of pedestrians at midblock crossings. This paper presents the results of research that analyzed driver yielding at 23 RRFB-enhanced midblock crossings on three-lane roadways with and without refuge medians islands in Oregon. The locations were chosen to represent a range of posted speed limits and average daily traffic that aligns with existing guidance for median and beacon installations. Sites were classified either as 1) no median refuge, RRFBs placed outside the roadway, 2) median refuge, RRFBs placed outside the roadway 3.) median refuge, RRFBs placed on the island and outside the roadway.  Yielding was determined following protocols established in prior research. Two hundred and seventy-six (276) hours of video footage was analyzed resulting in 3,065 crossing events (1,338 staged; 1,727 naturalistic) undertaken by 3,683 pedestrians. High yielding rates were observed—the average nearside yielding rate was 97%, with the lowest site having a rate of 89.9%. Yielding rates were generally higher on the farside. Due to sample size and consistently high yielding rates, it was not possible to make conclusive observations about the relationship between driver yielding and the presence of the median or additional beacons for the volume and speed combinations.  The data and analysis, however, generally indicate that the yielding rates increase with the addition of the median beacons. The findings also suggest the median refuge island with a beacon increases yielding. The increase in yielding was statistically significant at sites with 12,000-15,000 ADT.

MD MEHEDI HASAN, AECOM
Jun-Seok Oh, Western Michigan University
Valerian Kwigizile, Western Michigan University

Active transportation, especially walking, plays a vital role in ensuring a healthy and livable community. Walking, as a type of physical activity, helps in reducing the risk of health-related issues. Since the surrounding built environment around the walking site impacts walking experience, there is a need to investigate the influence of the physical and visual environmental attributes on pedestrian walking activity. In this study, we investigated the connection between human walking activities and surrounding physical and visual environment at a street level by applying artificial intelligence and spatial statistical analysis. A survey was initiated to collect a large amount of human trajectory data from Kalamazoo, Michigan through a mobile application developed by our research team. Google Street View (GSV) images were obtained to objectively assess the relationship between human trajectory data and visual environmental attributes. Statistical analysis results showed human walking activities were positively correlated with surrounding visual environmental attributes such as building enclosure, street light/pole, traffic sign and billboard, street greenery, and the enclosure of the sky view factor. In addition, we computed the visual environmental score (VES) along the sidewalks to assess the effect of visual environmental attributes on pedestrian walking experiences. The result showed that the VES index along the sidewalk segments varied by the purpose of pedestrian trip, but the VES index was positively correlated with the number of pedestrian trips.

Babak Mirbaha, Imam Khomeini International University
Nafise Fatemi Abhari, Imam Khomeini International University
AmirAbbas Rassafi, Imam Khomeini International University
Maziyar Layegh, Imam Khomeini International University

Child pedestrian injuries have become a major health challenge which needs preventive action. Children are vulnerable members of the society due to physical, emotional, and psychological characteristics and their low perception from the environment. The present study aimed to analyze the effect of the road environment, traffic conditions, and child pedestrian characteristics simultaneously in selecting gap acceptance among elementary students while crossing the urban crosswalks. In this regard, five crosswalks were selected in the vicinity of elementary schools in Qazvin, Iran. In addition, the required data were collected using the video images.  Post-Encroachment Time (PET) and Gap Time (GT) were considered as the criteria for assessing the risk. Further, binary logit models were used for analyzing the aspect of risk taking in elementary students. The results indicated that the factors such as age, gender, diagonal movement, suddenly turning back, vehicles and pedestrians’ volume, and the presence of speed bump play an important role in selecting the gap acceptance by students. In addition, male and female students revealed quite different behaviors while crossing the urban crosswalks. For this purpose, 18.18% of males put themselves at risky conditions while only 10.75% of females showed this kind of behavior. Furthermore, the mean walking speed of female was about 9% lower than that of for male students. Additionally, student's attention was higher when they were crossing the first half while they usually ignored the second half. Finally, the presence of speed bumps made students more self-confident when they were crossing.

Rahul Raoniar (rahul.raoniar@iitg.ac.in), Indian Institute of Technology, Guwahati
Akhilesh Maurya, Indian Institute of Technology, Guwahati

The current study primarily seeks to investigate the behavioural attribution of distraction in road crossing behaviour. An observational field study of 2360 pedestrians was conducted at three signalised intersections to compare the crossing behaviour for pedestrian-involved in distraction vs no distraction. The results revealed that about 28.7% of pedestrians were distracted while crossing the road. The young age pedestrians are observed to be more likely to talk on the mobile phone (18-29 years), text (<18 years) or hold a phone (18-29 years) in hand, while crossing through an intersection crosswalk. Pedestrians who text or eat/drink/smoke walks slowly compared to pedestrians with no distraction. Pedestrians texting on the mobile phone are 7.9% more likely to cross in “do not walk” signal phase. Pedestrian engaged in mobile phones while crossing pay less attention at traffic and surrounding before initiating crossing. Additionally, mobile phone talkers were observed to be 4.5% more likely to nearly hit/bump into another oncoming pedestrian compared to undistracted counterpart. Further, pedestrians crossing in a group (socially distracted) pay less attention to their surroundings, which might compromise pedestrian safety. The present study results may constitute vital information for planners and policymakers and play a pivotal role in identifying critical intersections and developing countermeasures to minimise the impact or occurrence of pedestrian distraction and unsafe behaviour.

Hong Deng, NanJing University of Science and Technology
Liu He, Nanjing University
Tangyi Guo, Nanjing University
Kun Tang, Nanjing University

In order to meet the global common needs of COVID-19 epidemic containment, it is particularly important to establish a pedestrian walking model considering the interaction between pedestrians and epidemic situation, aiming at the problem of massive epidemic spread caused by pedestrian aggregation in public places. In this paper, based on the social force model, we established the pedestrian movement model under the epidemic background firstly. Then, according to the characteristics of COVID-19 epidemic, an infection model is established to simulate the single person's dynamic behavior of infection, thus, the pedestrian walking model with population epidemic coupling is proposed. Finally, the spread law of the epidemic situation was simulated by using computer simulation technology. The results showed that pedestrian flow rate, coverage rate of pedestrian protection measures, pedestrian spacing and pedestrian walking speed were the key factors affecting the spread of the epidemic. When the pedestrian flow rate was 1700 person / h, the number of infected persons within 500s had reached 27. Therefore, the control of pedestrian flow rate can be the most important direction of epidemic containment.

Khaled Shaaban (kshaaban@uvu.edu), Utah Valley University
Deepty Muley, Qatar University
Abdulla Mohammed, Qatar University

Jaywalking is a problem that requires serious consideration especially in developing countries, which are facing the impact of increasing population and relatively fewer resources. Planning deficiencies at times and poor enforcement cause this type of illegal crossing pattern. The purpose of this study is to investigate such maneuvers by analyzing pedestrians’ ways and speeds of illegally crossing a major urban road in the city of Doha, Qatar. The crossing speeds were determined for two cases: 1794 pedestrians crossing without conflict and 972 pedestrians crossing with conflict. The results indicated that the crossing speeds of pedestrians crossing without conflict were much lower than crossing speeds of pedestrians crossing with conflict. The results suggested that the difference in mean crossing speed for gender, age group, type of clothes, size of group, pedestrian path, flow with the pedestrian, and waiting time were statistically significant, and the difference in the mean speed was not statistically significant for carrying luggage and flow against the pedestrian for both cases. A stepwise regression analysis was conducted for both cases to determine significant variables predicting the crossing speeds. These values can be used in future studies as an input in gap acceptance analysis and pedestrian simulation studies. Also, understanding the characteristics of this type of maneuver can be useful for policy-makers in this region to develop strategies and to target the high-risk categories identified in the study.

Nick Ferenchak (ferenchak@unm.edu), University of New Mexico
Sheheryar Shafique, University of New Mexico

The objective of this work was to better understand pedestrians’ understanding, trust, comfort, and acceptance of autonomous vehicles (AVs) by testing AV external human-machine interfaces (eHMIs). AV developers have generated a variety of eHMI concepts to accomplish AV-to-human communication, but little research exists regarding the most effective approach. Using a within-subject experiment design, 47 participants interacted with AVs in a virtual reality environment. The AVs possessed a variety of eHMIs based on those designed and proposed by AV developers. We administered a Likert scale survey to measure participants’ perceptions of the eHMIs. We analyzed responses with ordinal logistic regressions accounting for gender, participants’ stated interest in AVs, and either eHMI presence, eHMI design, or yielding behavior. The presence of an eHMI was found to improve participants’ perceptions of AVs. Although females generally reported higher levels of understanding, trust, comfort, and acceptance, males’ scores increased more significantly with the introduction of an eHMI. Text eHMIs outperformed non-textual interfaces, with participants noting the best perceptions with the text eHMI located on the AV’s grille. Understanding, trust, and comfort followed a similar pattern (these metrics were greatly improved with the presence of an eHMI and best outcomes were experienced with text displays), but acceptance did not (this metric had a small response to the presence of an eHMI and the best outcomes were with the LED windshield). This research supports the development of a standard, uniform AV-pedestrian communication strategy and strengthens the connection between humans and AVs.

Fan Zuo (fz380@nyu.edu), New York University
Jingqin Gao, New York University
Abdullah Kurkcu, Ulteig
Hong Yang, Old Dominion University
Kaan Ozbay, New York University
Qingyu Ma, Old Dominion University

The fast-evolving COVID-19 pandemic has dramatically reshaped urban travel patterns. Through the pandemic, one term that nearly everyone throughout the world has quickly become familiar with is "social distancing." In this research, we explore the relationship between social distancing and urban mobility during the pandemic. Understanding the social distancing behavior of people will allow urban planners and developers to better understand the new norm of urban mobility amid the pandemic and what it might hold for individual mobility post-pandemic or in future pandemics. This paper aims to shorten that gap by developing a data-driven analytical framework that leverages existing public video data sources and advanced computer vision techniques to monitor the evolution of social distancing patterns in urban areas. Specifically, the proposed framework develops a deep-learning approach with a pre-trained convolutional neural network to mine the massive amount of public video data captured in urban areas. Real-time traffic camera data collected in New York City was used to demonstrate the feasibility and validity in using the proposed approach to analyze pedestrian social distancing patterns. The results show that microscopic pedestrian social distancing patterns can be quantified by the proposed real-distance approximation method. The estimated social distances among individuals can be compared to social distancing guidelines to evaluate policy compliance/effectiveness. The quantified social distancing status will aid decision-makers and the public to have a better understanding of the prevailing social contact challenges. It also provides informative insights into the development of response strategies for different reopening phases and potential future scenarios.

Laura Harris, University of Tennessee, Knoxville
Russell Graves, University of Tennessee, Knoxville
Ramin Arvin, University of Tennessee, Knoxville
Asad J. Khattak, University of Tennessee
Subhadeep Chakraborty, University of Tennessee

Nearly 6000 pedestrian deaths occur every year in the US and many more are injured in collisions with vehicles. To improve safety of pedestrians, the goal of this study is to determine if crossing choice and jaywalk crossing behaviors could be successfully understood from data collected from participants in an immersive virtual reality simulation. After designing the experiment and collecting data from subjects, rigorous models were estimated. A Heckman selection model was used to analyze the data for a safe crossing parameter, time-to-collision, and crossing choice. The model was chosen to account for whether a crosswalk was used or the pedestrian decided to jaywalk. The study explored predictors for both crossing choice and crossing behavior of time-to-collision. Additionally, alerts were given to pedestrians to examine the impact of vehicle-to-pedestrian communication. The alert was designed to encourage a safer crossing by playing a set of natural sounds when a crossing attempt would result in a time-to-collision less than a given threshold. The alert was not directly found to be a significant predictor of higher time-to-collision. However, scenarios with the alert were found to have a significantly higher delay and time-to-collision than the baseline scenario. Another explanation is that unobserved factors or the strength of the alert could have influenced the outcome. The results are in agreement with cited literature and provides further avenues for study. Furthermore, we have developed a virtual reality simulation which allows us to collect pedestrian behaviors and test novel vehicle-to-pedestrian communications.

Jie Gao (j.gao0701@hotmail.com), Chang'an University
Toshiyuki Yamamoto, Nagoya University
Marco Helbich, Universiteit Utrecht

This study examined 1) socio-demographic moderators on the associations between built environment and 2 transit-related walking; and 2) the extent to which these relationships and socio-demographic moderating 3 effects are explained by the mediator, car ownership. Our sample was obtained from the Dutch National 4 Travel Survey 2010-2014 including 92,298 people aged ≥18 years. Mediation and mediated moderation 5 analyses were performed via Vector Generalized Linear Models (VGLMs). Our results showed that car 6 ownership mediated the associations of address density with transit-related walking. Mediated moderation 7 showed car ownership had effects on the associations of address density, crossing density, number of bus 8 stops, distance to the public transport and daily facilitates, and the percentage of green space with transit-9 related walking. Further, our results revealed that densely populated neighborhoods with access to public 10 transport have the potential to significantly contribute to transit-related walking. Car ownership should be 11 considered as a mediated variable to correctly determine the usefulness of urban planning policies which 12 intend to discourage car use.

Abdullah Alhawsawi, University of Melbourne
Majid Sarvi, University of Melbourne
Abbas Rajabifard, University of Melbourne
Jianyu Wang, University of Melbourne

Gaining an understanding of the collective movement of humans and ascertaining how individuals interact within their physical environment has attracted the attention of many researchers in the crowd dynamics field. We conducted an experimental study to observe the interaction of collective motion among people and to understand the characteristics of pedestrians when passing obstacles of different sizes (bar-shaped, 1.2m, 2.4m, 3.6m and 4.8m) and one narrow exit and three different flow rates in walking and running conditions. Our results indicated that, in the collision avoidance section, pedestrians reacted early to the obstacles and changed the direction of their walking by quickly manoeuvering to left or right. In terms of the pedestrians’ speed, while performing these tasks, the average velocity was significantly impacted, showing decreases as the obstacle size increased, meaning that different obstacle size will affect different flow and speed levels. In term of the travel time, passage time was shorter when participants were in the medium flow rate experiments, indicating the optimum layout for the evacuation efficiency. Results also indicated that lateral distance when passing the obstacle averaged 0.3m to 0.7m, depending on the flow rate and speed levels. Finally, the body swaying part was explored. It was observed that the average body swaying was smaller in the high-speed regime compared to the low-speed regime except for (HF & 4.8m) experiment. The results of these studies are expected to provide an insight into the characteristics of pedestrian behaviour when avoiding objects and could help enhance the  agent-based models.

Shile Zhang (shirleyzhang@Knights.ucf.edu), University of Central Florida
Mohamed Abdel-Aty, University of Central Florida
Yina Wu, University of Central Florida
Ou Zheng, University of Central Florida

Pedestrians’ red-light crossing behaviors can pose a threat to traffic safety. This study predicts pedestrians’ red-light crossing intentions using various machine learning models. Pose estimation is used to generate pedestrians' key points from video data. Some motion features of pedestrians are extracted and transformed. Pedestrians’ red-light crossing intentions are labeled with signal timing data and a crossing intention model. Four machine learning models, Support Vector Machine (SVM), Random Forest (RF), Gradient Boosting (GBM), and eXtreme Gradient Boosting (XGBoost) are used to predict pedestrians’ red-light crossing intentions. The best model achieves an accuracy of 89.1% at one intersection. This model can be further implemented in the context of vehicle-to-pedestrian (V2P) applications and thus preventing crashes due to the pedestrians’ red-light crossing behaviors.

Indrajit Ghosh, Indian Institute of Technology, Roorkee
Abhinav Kumar, National Institute of Technology, Uttarakhand
APURWA DHOKE, National Institute of Technology, Calicut

With the rapid urbanization of geographical spaces all around the world, the safety of pedestrians is a major concern on urban roads. In developing economies like India, an unprecedented increase in pedestrian crashes has been observed at the intersections. The present study focuses on pedestrian violations behavior while crossing at the signalized intersections. With the help of hazard-based duration models, the waiting duration of red-light violators has been analyzed. In addition, the response time of pedestrians during conflict has also been modeled with the help of a hazard-based duration approach. Four signalized intersections from Nagpur City in India have been selected for the survival analysis. Kaplan Meier survival curves have been plotted for both the duration, i.e., waiting time and response time. With the help of the semi-parametric Cox proportional hazard model, the various factors have been identified that are related to the survival probability of the pedestrians during crossings. However, the model results were not found satisfactory since the explanatory variables failed in the proportional hazard assumption. Therefore, the parametric accelerated failure time (AFT)  model was utilized to determine the various covariates that affected the waiting time and the response time. The Weibull model was found to be the best fit for waiting duration analysis, while the Log-Logistic model was considered for the study of response time. The developed models can help to understand the external factors as well as personal features of the pedestrians in relation to the risk involved during violation crossings.

Juliane Stark (juliane.stark@boku.ac.at), University of Natural Resources and Life Sciences
Katja Ruzsicska, University of Natural Resources and Life Sciences
Thomas Wiesmann, Universitat fur Bodenkultur, Wien

In Austria, distraction and carelessness are the number one cause of accidents. This affects all types of traffic. The use of smartphones makes a significant contribution to distraction in road traffic. While the issue of car drivers and adult pedestrians has already been investigated in numerous publications, the extent of the problem and its impact on children and adolescents walking on foot is hardly known. Within the framework of an explorative study, almost 2,800 crossing events of school children in front of an educational center in the city of Vienna (Austria) were observed and analyzed. The results show that 44% of the pupils observed were engaged in some kind of use or were visibly holding a mobile phone in their hands. In a subsequent qualitative survey, it was observed that especially when the gaze is directed to the device, waiting time before crossing the road as well as the crossing time is longer and attention is turned away from traffic for longer periods of time. The results underline the need for training to educate children and adolescents about safe behavior in road traffic.