Porto J., Roussou S., Andrade M., Ziakopoulos A., Yannis G., “Conflict detection and analysis in urban arterial roads of Brasília, Brazil, using HD-CCTV monitoring cameras and the YOLO model”, Proceedings of the 10th Road Safety and Simulation International Conference, Naples, 23-26 June 2026.

Traffic conflict analysis has been advocated in the past years as a surrogate measures of safety, complementing or even substituting crash-based assessments by enabling proactive evaluation of road safety. Although traffic conflicts occur more frequently than crashes, the require formal definition through indicators, either based on individual movements (e.g. harsh braking or harsh swerving) or on spatiotemporal proximity measures. This work evaluates the feasibility of automated traffic conflict detection at four major urban intersections in Brasília, Federal District, Brazil, using pre-existing surveillance footage. Video data were collected with HD-CCTV cameras from the regional monitoring system. Motorized road users were detected using YOLOv11m, while object tracking was implemented in Python following state-of-the-art approaches from the literature to estimate frame-by-frame trajectories. These trajectories were subsequently vectorized and used to calculate Time to Collision (TTC) between pairs of vehicles. The results demonstrate a substantial improvement over previous automated safety monitoring efforts conducted in the same area. However, steep camera angles, limited image quality, and the erratic motion of motorcyclists still hinder robust tracking performance. The tracking pipeline performs reliably for buses and passenger cars, although occasional duplicate detections occur when temporal matching is disrupted. In contrast, motorcyclists are not consistently detected under real-world conditions, and trucks tend to be over detected. Despite these limitations, the conflict detection framework provides meaningful insights into interactions among road users and highlights the effectiveness of complete traffic flow separation in reducing hazardous encounters, even in the presence of motorcycles. Future work will focus on improving object matching criteria to enhance trajectory continuity and prediction stability. More broadly, this research demonstrates the potential of integrating tracking algorithms, surveillance systems, and surrogate safety measures for scalable and continuous monitoring of urban road safety.