A Study of Vehicle-Pedestrian Jay-walk Interactions at Median Road Locations in Bangladesh: Using Image Recognition And Image Processing Techniques

Abstract

Pedestrian safety remains a global concern as a significant number of traffic-related injuries and fatalities involve pedestrians. Factors such as uncontrolled traffic flow, mid-block crossings, and inadequate road safety infrastructure exacerbate this issue. According to the Federal Highway Administration (FHWA), these elements are particularly prevalent in densely populated and poorly regulated urban environments. In Bangladesh, the capital city of Dhaka consistently reports the highest traffic fatality rates, underscoring the urgency of addressing pedestrian safety in developing countries. Although developed nations like the United States, Ireland, and Greece have conducted extensive studies on jaywalking in controlled traffic scenarios, limited research has focused on underdeveloped countries, where traffic is often mixed and unpredictable, making the problem more complex. This study investigates the influence of the distance and speed of oncoming vehicles on jaywalking decisions across three urban zone types: industrial, residential, and commercial. These zones were chosen for their diverse traffic characteristics and pedestrian usage patterns. Dashcam footage from vehicles traveling in these areas was used to collect data. The selected locations include Bangla-motor and Gulshan (commercial zones), Mirpur and Banani (residential zones), and Gazipur and Tejgaon Industrial Area (industrial zones). Advanced image recognition techniques, specifically the YOLOv8 (You Only Look Once) machine learning model, were employed to detect jaywalking behavior. The model also measured the speed and distance of approaching vehicles, providing precise and real-time data. To analyze jaywalking patterns, Kernel Density Estimation (KDE) plots were generated, allowing a visual understanding of pedestrian crossing behavior. Statistical analyses, including two-sample t-tests for both equal and unequal variance, were used to compare the xiii speed and distance of oncoming vehicles across different zones and timeframes, specifically during morning, evening, and off-peak hours. The findings reveal significant variations in pedestrian and vehicular behavior based on the zone and time of day. Industrial zones recorded the highest vehicle speeds during weekdays, averaging 16.41 km/h, reflecting the urgency and high volume of traffic in these areas. Conversely, residential zones exhibited the highest vehicle speeds on weekends, averaging 26.18 km/h, likely due to reduced weekday congestion and higher recreational movement. Regarding crossing distances, pedestrians in residential zones maintained the longest distance from oncoming vehicles during off-peak hours on weekdays (7.48 m). On weekends, the residential zones also showed the greatest crossing distance during morning peak hours (8.05 m), suggesting a more cautious approach by pedestrians during specific periods. This study provides valuable insights for urban planners and policymakers aiming to enhance pedestrian safety. The detailed behavioral patterns observed across different zones and timeframes highlight the need for tailored interventions. Recommendations include the installation of safer pedestrian crossings, implementation of traffic-calming measures, stricter enforcement of jaywalking laws, and the integration of intelligent traffic management systems. Additionally, improving road safety infrastructure in high-risk areas such as industrial zones could significantly reduce pedestrian-vehicle conflicts. These findings serve as a foundation for developing comprehensive road safety strategies not only in Bangladesh but also in other developing countries facing similar challenges.