A Comprehensive Survey on Multi-Object Tracking in Video using Artificial Intelligence

Abstract

Multi-object tracking (MOT) is an important computer vision job with several applications in robotics, autonomous navigation, traffic monitoring, and surveillance. A complete real-time MOT framework, backed by TensorFlow optimization and Zusland deployment tools, is presented in this study. It uses YOLOv5 for fast object recognition and DeepSORT for identity-preserving tracking. The suggested solution successfully addresses key issues, including occlusion, identity switches, congested surroundings, and latency limits, guaranteeing scalability and resilience across a variety of video streams. Experimental assessments show good performance in real-time feasibility, tracking accuracy, and precision, with few identity shifts attaining 42 FPS, 87.5% MOTA, and 81.2% MOTP. This study lays the groundwork for future developments in AI-driven tracking systems by enabling a modular, deployable, and scalable solution for real-world video analytics by bridging the gap between scholarly research and real-world implementation.