File(s) under embargo
Reason: We ned to do further research based on current work
until file(s) become available
AN APPLICATION OF UNMANNED AERIAL VEHICLES IN INTERSECTION TRAFFIC MONITORING
Motor vehicle crashes at represent a major cause of fatalities and injuries. As such, road transportation agencies continue to seek proactive intersection traffic monitoring initiatives to reduce crashes. Proper monitoring and assessment of crash risk can not only help identify and mitigate safety hazards in real time but also enhance the design of safer facilities and promote safe road-user behavior. It has been suggested in recent literature that unmanned aerial vehicles (UAVs), given their wide visual field and movement flexibility, can potentially help monitor road traffic operations when deployed on the field. In addition, it is recognized that it is still challenging to realize a large-scale ground-based vehicle-to-everything (V2X) network at the current time and in the very near future. In this regards, UAVs can play a critical connectivity role by serving as a hub to facilitate communications among roadway entities (vehicles, infrastructure, and pedestrians). This thesis first presented a methodology that integrates UAVs and V2X connectivity to track the trajectory of intersection users and to monitor potential collisions at intersections. The proposed methodology includes deep-learning-based tracking algorithms and time-to-collision assessments. The methodology was applied using a case study, and the results demonstrated the efficacy of the tracking methodology. Next, the thesis addressed the issue of image quality. During inclement weather, traffic monitoring can be challenging because the video quality is often corrupted by streaks of falling rain on the video image. This may hinder the reliability of characterizing the road environment and road-user behavior during such events. To fully exploit the benefits of video captured by UAVs in traffic monitoring, crash risk assessment, and other safety-related domains, it is critical to ensure high video quality. Therefore, this thesis proposed a two-stage self-supervised learning method to remove rain streaks in traffic videos, where the first and second stages address intra- and inter-frame noise, respectively. The results suggest that the proposed method provides satisfactory performance.