一种改进TLD的目标跟踪算法  

Study on an improved TLD algorithm for target tracking

作　　者：王铁东 任世卿[1] Wang Tiedong;Ren Shiqing(Information Science and Engineering School,Shenyang Ligong University,Shenyang 110159,China)

机构地区：[1]沈阳理工大学信息科学与工程学院,辽宁沈阳110159

出　　处：《江苏科技信息》2018年第1期52-53,56,共3页Jiangsu Science and Technology Information

摘　　要：目标跟踪是模式识别领域传统且重要的分支。传统跟踪算法有背景差分法、帧间差分法、卡尔曼滤波等,而TLD算法从检测和跟踪两个维度,将检测与跟踪的结果同时作用于跟踪,在跟踪任务中展现出较好性能。同时TLD算法的跟踪模块采取了时间开销较大的光流法。文章针对时间开销较大的不足,提出一种改进TLD的目标跟踪算法,将光流法代替为时间开销较小的meanshift算法。实验结果表明,该方法能够有效地减少时间开销,提高运动目标跟踪效率。Target tracking is a traditional and important branch of pattern recognition.Frame difference,background difference,Calman filtering are among the category of the traditional target tracking algorithm,the TLD simultaneously applies the detection and tracking results to the tracking,which improves the performance of tracking.Meanwhile,the tracking module of TLD algorithm takes the optical flow method which cost a lot of time.In order to solve this problem,this article proposes a target tracking algorithm to improve TLD,replacing the optical flow method with a meanshift algorithm which consumes less time.Experimental results show that this method can effectively reduce the time consumed and improve the tracking efficiency of moving targets.

关 键 词：运动目标跟踪 模式识别 检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]