RGB-D与MeanShift相结合的实时目标跟踪  

Real-time object tracking combined RGB-D with MeanShift

作　　者：王广龙 田杰 朱文杰 方丹 WANG Guanglong;TIAN Jie;ZHU Wenjie;FANG Dan(Laboratory of Nanotechnology and Microsystems,Shijiazhuang Campus of Army Engineering University,Shijiazhuang 050003,China)

机构地区：[1]陆军工程大学石家庄校区纳米技术与微系统实验室

出　　处：《中南大学学报（自然科学版）》2019年第9期2163-2170,共8页Journal of Central South University:Science and Technology

基　　金：国防预研基金资助项目(51327030104)~~

摘　　要：针对MeanShift跟踪算法目标易受到周围相似背景区域干扰的问题,引入图像的RGB-D特征,研究并实现结合RGB-D与MeanShift的实时目标跟踪算法。首先,提出一种新颖的相似性度量方法优化二次型距离(advanced quadratic-form distance,AQFD);其次,基于颜色和深度特征完善了特征权重在线自适应调整机制和目标模型更新策略;最后,对该算法进行仿真分析并基于Jetson TX2平台实现。研究结果表明:本文算法在目标受到周围相似背景区域干扰情况下仍能够完成目标跟踪任务,优于先进算法Staple,具有良好的准确性与鲁棒性,并且能够应用于实时跟踪。Arm to solve the problem that the object is susceptible to interference from similar surrounding background in MeanShift,RGB-D feature was introduced,and thus a real-time object tracking combined RGB-D with MeanShift was proposed.Firstly,a novel similarity measurement method,advanced quadratic-form distance(AQFD),was put forward.Secondly,the online self-adaptive adjustment mechanism of feature weights and updating strategy of object model were improved.Finally,the method was simulated based on Matlab and implemented on Jetson TX2.The results show that the proposed method which is better than Staple performs well in accuracy and robustness in the case that the object is disturbed by similar surrounding background,and it can be applied to real-time tracking applications.

关 键 词：目标跟踪 MEANSHIFT RGB-D 实时跟踪 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]