Research on Optimized Vehicle Detection and Tracking Algorithms for Intelligent Transportation Based on Multi-Object Tracking

In the context of ongoing smart city development, intelligent transportation has become a core component of urban governance and traffic coordination. This study designs a vehicle dynamic perception algorithm for road traffic scenarios, which employs an object detection module for automatic vehicle identification within road traffic scenes and integrates an object tracking module for continuous monitoring of vehicle motion states, thereby effectively improving traffic management efficiency and recognition accuracy. Leveraging deep learning and computer vision methods, the constructed system can efficiently and accurately distinguish various types of vehicles and their operational states in road traffic scenes, significantly improving road traffic throughput while reducing the need for manual intervention. Furthermore, this paper proposes a dynamic multi-object tracking scheme based on a multi-constraint composite perception association filtering mechanism, which constructs multiple constraint conditions by introducing spatial position and heading angle information of moving targets, and designs a multi-constraint generalized probabilistic data association filtering method, achieving robust measurement association and state estimation under single-sensor conditions in complex dynamic scenarios such as multi-target intersection and crossing. Experimental validation demonstrates that the proposed method surpasses existing conventional approaches in both vehicle identification accuracy and real-time response capability. Additionally, to address the increased tracking difficulty when target motion exhibits nonlinear characteristics, this paper introduces Long Short-Term Memory networks (LSTM) into the ByteTrack framework for improvement. Comparative results show that the improved ByteTrack model achieves a 5.0% increase in tracking accuracy, an 8.7% improvement in processing frame rate, and an 21.9% reduction in identity switch frequency, with significantly enhanced tracking robustness under complex conditions.