Robust Multi-Camera Tracking with YOLOv10 and OSNet: Improving Accuracy through DeepSort and ID ReAssignment

This paper proposes a robust solution for multi-video object detection, tracking, and re-identification using the latest advancements in deep learning models, specifically YOLOv10 for object detection, DeepSORT for tracking, and Omni-Scale ReID (OSNet) for person re-identification.Our approach addresses the challenge of efficiently integrating multiple video streams into a single display while ensuring accurate detection and tracking across frames.The proposed system processes four simultaneous video feeds, presented in a 2x2 grid format, offering a unified view for real-time monitoring.YOLOv10's anchor-free architecture and attention mechanisms enhance detection accuracy and speed, while DeepSORT provides consistent tracking through its combination of motion and appearance models.Omni-Scale ReID further ensures robust re-identification of individuals across video streams, accommodating scale variations and environmental changes.This comprehensive framework is optimized for applications such as surveillance, traffic monitoring, and smart environments, offering a scalable, real-time solution with high accuracy and resilience to occlusion and dynamic conditions.Extensive experiments demonstrate the system's efficacy in handling complex multi-video scenarios, outperforming previous methodologies in detection and re-identification tasks.