基于多摄像头的实时人物追踪系统

Real-Time Person Tracking System Based on Multiple Cameras

摘要: 在智慧城市与智能安防快速发展的背景下，多摄像头人物追踪技术成为提升公共安全管理效率的核心需求。传统方案存在跨摄像头身份关联失效、计算资源消耗大及实时性不足等问题。本文提出一种基于边缘计算与中心数据库协同架构的实时人物追踪系统，采用轻量化模型YOLOv5s进行人体检测，结合OSNet_x0.25提取512维特征向量，通过向量数据库实现毫秒级特征匹配。实验表明，该系统在Market-1501数据集上Rank-1准确率达86.7%，640 × 640视频流实时处理帧率为8.2 FPS，在校园场景中验证了工程实用性。

Abstract: In the context of the rapid development of smart cities and intelligent security, multi-camera person tracking technology has become a core requirement for improving the efficiency of public safety management. Traditional solutions suffer from issues such as failed cross-camera identity association, high computational resource consumption, and insufficient real-time performance. This paper proposes a real-time person tracking system based on a collaborative architecture of edge computing and a central database. It employs the lightweight YOLOv5s model for human detection, combines it with OSNet_x0.25 to extract 512-dimensional feature vectors, and achieves millisecond-level feature matching through a vector database. Experiments show that the system achieves a Rank-1 accuracy of 86.7% on the Market-1501 dataset, with a real-time processing frame rate of 8.2 FPS for 640 × 640 video streams, validating its engineering practicality in campus scenarios.