Research on Robot Object Detection and Tracking Based on Deep Learning Algorithms

With the increasing demands for accuracy in object detection and tracking in fields such as intelligent surveillance and autonomous driving, traditional multi-object tracking algorithms still face challenges such as identity matching instability in complex scenarios like deformation, reappearance of targets, and fast motion. This paper aims to optimize the YOLOv5 algorithm and design a multi-object tracking architecture that integrates the YOLOv5 detector with a ReID module. This integration enhances tracking performance in complex scenarios involving deformation, scale variation, and fast motion. Experiments were conducted on the OTB2015 dataset to compare the performance of the proposed optimized YOLOv5 algorithm with mainstream algorithms such as YOLOv5, TransT, Ocean, and SiamR-CNN. The experimental results demonstrate that the optimized algorithm, which combines YOLOv5 with ReID, outperforms other algorithms in typical challenging scenarios such as deformation, scale variation, and fast motion. It also shows comprehensive improvements in metrics such as accuracy, precision, recall, and F1-score. The main innovation of this paper lies in the tight integration of the optimized YOLOv5 with the ReID feature extraction module, achieving efficient collaboration between detection and identification, and showing strong practical application potential.