基于多特征结合的长时跟踪算法设计  

Design of long-term tracking algorithm based on multi-feature combination

作　　者：曾君怀 宋亚男[1] 徐荣华[1] 黎安民 Zeng Junhuai;Song Yanan;Xu Ronghua;Li Anmin(Guangdong University of Technology,Guangzhou 510006,China)

机构地区：[1]广东工业大学,广东广州510006

出　　处：《无线互联科技》2021年第17期58-60,共3页Wireless Internet Technology

基　　金：广东工业大学青年基金,项目编号:18ZK0021。

摘　　要：文章针对目前多数长时跟踪算法对同类物体区分能力较弱且在复杂背景下跟踪不理想的问题,结合运动和外观特征的长时跟踪算法进行设计。外观特征使用了重识别算法,在同类物体区分能力较好。运动特征借鉴了SORT跟踪算法原理来实现。主要分3个模块论述了设计的长时跟踪算法框架,最后就设计实验算法与其他两种长时跟踪算法进行了比对。In order to solve the problem that most long-term tracking algorithms have weak ability to distinguish the same kind of objects and are not ideal in complex background.In this paper,a long-term tracking algorithm combining motion and appearance features is designed.Appearance features use re recognition algorithm.The algorithm has good ability to distinguish similar objects.The motion feature is realized by using the SORT algorithm.This paper is divided into three modules to discuss the framework of long-time tracking algorithm.Finally,experiments are designed to compare this algorithm with the other two long-time tracking algorithms.

关 键 词：长时跟踪 运动特征 外观特征 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]