基于多特征融合的多目标跟踪算法研究

Research on Multi-object Tracking Algorithm based on Multiple Features

视频监控技术在很大程度上为人们的出行安全提供了保障,当前以人工监控居多,随着数据量越来越大,场景越来越复杂,人工监控的弊端也日益凸显,智能监控系统应运而生。目标检测与目标跟踪是一套智能监控中最重要的两个部分,两个任务正在成为一个热门的研究问题。以FoveaBox检测算法为基础,通过多组对比实验确定损失函数及关联方案,设计了一个融合了多种特征的多目标行人跟踪系统,实现了检测与数据关联合并的一步跟踪,同时为该领域提供了一个研究方向。与其他算法相比,提出的MOTA和MT指标均优于其他算法。

Video surveillance has provided a guarantee for people’s travel safety to a large degree.Currently,manual surveillance is mostly used.However,as the amount of data becomes larger and the scene more sophisticated,the disadvantages of relying on manual surveillance is increasing and intelligent surveillance systems emerged.Target detection and target tracking are the two most important parts of a set of intelligent monitoring,and the two tasks are becoming a hot research problem.This paper uses multiple sets of comparative experiments to determine the loss function and correlation scheme,then designs a multi-object pedestrian tracking system that integrates multiple features,which integrated detection and data association to realize one-step tracking,and also provides a research direction for the field.Compared with other algorithms,the proposed MOTA and MT indicators are better than other algorithms.