برآورد شاخص‌های ترافیکی تقاطع چراغ‌دار مبتنی بر تصاویر پهپاد

In response to the necessity of traffic analysis at oversaturated intersections with non-lane-based vehicle behavior, this study proposes a framework comprising pre-processing, frame stabilization, georeferencing, and vehicle detection and tracking for the automated estimation of intersection performance measures. To this end, vehicle detection and tracking were performed on imagery from the Tohid intersection in Tehran using the YOLOv8 model and the ByteTrack algorithm, respectively. A mean Average Precision (mAP@0.5) of 0.961 for the detection model indicated its high accuracy. Subsequently, intersection performance indicators were calculated and analyzed based on the extracted trajectories and field data. The analyses reveal a functional imbalance across the intersection arms; specifically, the East and West arms face discharge challenges, recording low average exit speeds (15 and 22 km/h, respectively) and limited entry saturation flow rates per unit width (327 and 317 veh/h/m). In contrast, the North and South arms demonstrated better performance, with higher exit speeds (32 and 23 km/h, respectively) and adequate entry saturation flow rates per unit width (567 and 535 veh/h/m). However, delays exceeding 80 seconds at all entries and the degradation of the overall intersection Level of Service (LOS) to grade F indicate critical conditions and the necessity to revise the geometric design or signal timing. The findings of this study demonstrate that the proposed framework provides an efficient platform for estimating traffic variables and analyzing the performance of signalized intersections.