ميلاد بيات

Extracting traffic parameters is a cornerstone of traffic analysis an‎d the design of intelligent transportation systems. However, the current state of this process requires improvement to enable the deployment of a comprehensive, accurate, efficient, reliable, an‎d sustainable system with real-time performance. To this end, the present study aims to design an advanced an‎d efficient system for extracting traffic parameters by performing real-time vehicle detection an‎d tracking within surveillance systems based on video analytics. This computer-vision approach is implemented as an integrated pipeline comprising detection, tracking, time-dependent trajectory processing, an‎d the extraction of the target data. The proposed system employs YOLOv10 for object detection an‎d ByteTrack for tracking, together with transfer learning, to strike an optimal balance between performance metrics an‎d processing time while recovering vehicle trajectories. Finally, by applying time-dependent trajectory processing an‎d coordinate conversion via a perspective transform, the desired traffic parameters are obtained. eva‎luation on the BrnoCompSpeed dataset shows that the proposed system achieves overall simple an‎d weighted mean accuracies of 98.54% an‎d 96.84%, respectively, in estimating vehicle counts, an‎d correctly identifies an‎d records all lane-change events. Moreover, for vehicle-speed estimation, it attains mean absolute an‎d relative errors of 1.84 km/h an‎d 2.32%, respectively. These results indicate that the proposed system can estimate the total number of vehicles an‎d lane-change events with desirable accuracy, an‎d they demonstrate the system’s relative stability an‎d capability in estimating vehicle speeds such that, in terms of reliability an‎d predictability, it is judged suitable for field deployment.

استخراج پارامترهاي ترافيكي , بينايي كامپيوتري , تشخيص و رديابي بلادرنگ اشيا , سامانه‌هاي حمل‌ونقل هوشمند , يادگيري عميق

Computer Vision , Deep Learning , Intelligent Transportation Systems (ITS) , Real-Time Object Detection & Tracking , Traffic Parameter Extraction

Milad Bayat

Ali Tavakoli Kashani - Mahmood Ahmadi Nejad