An Enhanced Multiple Object Tracking Method With Improved Feature Fusion and Cross-Camera Trajectory Matching for Closed Scenes

Abstract

Multi-Object Tracking (MOT) is a task in computer vision. It aims to detect and track objects across video frames while keeping their identities consistent. Traditional surveillance systems mainly record video footage without automated object identification or tracking. This requires monitoring and leads to inefficiency in crowded environments. This project proposes a multi-object tracking system. It integrates YOLOv8 for real-time object detection, DeepSORT for tracking and face recognition for identity verification. The system detects persons and vehicles from real-time video streams or recorded footage. It assigns tracking IDs to maintain object identity across frames. A global ID management module ensures that the same individual receives the identity when reappearing in different frames or cameras. Optimization techniques like frame skipping and GPU acceleration improve processing speed and system efficiency. Experimental evaluation demonstrates object detection, consistent tracking performance and reduced identity switching.The proposed system is suitable for surveillance applications like public security monitoring, traffic analysis and crowd management.