Vehicle Detection and Multi-Target Tracking System in Intelligent Transportation using YOLOv8 and Adaptive Deep Sorting Algorithm

With the growing complexity of urban traffic, traditional vehicle detection and tracking methods struggle to balance accuracy and real-time performance in dynamic, high-density scenarios. This study proposes an intelligent transportation system integrating YOLOv8 for detection and an adaptive DeepSORT algorithm for multi-target tracking. YOLOv8, enhanced with data augmentation and scenario-specific training, improves vehicle recognition across diverse environments. The adaptive DeepSORT introduces dynamic Kalman filter updates, appearance feature weight adjustments, and a re-identification mechanism. Experiments on UA-DETRAC and KITTI datasets reveal that in low-density traffic, the system achieves a 92.3% mAP detection accuracy and 97.8% tracking success rate with 32.5 FPS. However, in high-density scenarios, detection accuracy declines to 84.6%, and ID switches increase to 24 instances, while frame rate drops to 18.3 FPS. The adaptive mechanisms mitigate tracking mismatches, maintaining robustness despite performance degradation. These findings demonstrate the system’s practicality for real-world deployment, though challenges persist in dense traffic.