基于自适应压缩跟踪算法的视觉目标跟踪  

Visual Target Tracking Based on Adaptive Compression Tracking Algorithm

作　　者：裴春梅[1] 胡亦[1] 朱恭生[1] 景妮琴[1] PEI Chun-mei;HU Yi;ZHU Gong-sheng;JING Ni-qin(Department of Electronic Technology ,Beijing Polytechnic ,Beijing 100176,China)

机构地区：[1]北京电子科技职业学院电子技术系,北京100176

出　　处：《内蒙古师范大学学报（自然科学汉文版）》2018年第2期132-136,140,共6页Journal of Inner Mongolia Normal University(Natural Science Edition)

基　　金：西藏自治区自然科学基金课题(2015ZR-14-58)

摘　　要：针对压缩跟踪算法在目标发生遮挡、快速移动、有相似目标情况存在跟踪漂移的问题,提出了基于卡尔曼滤波的自适应学习压缩跟踪算法.该算法首先利用压缩跟踪算法对目标进行定位,然后根据跟踪结果的置信图对分类器参数自适应更新,当判定目标严重遮挡时,利用卡尔曼滤波进行预测估计.实验结果表明,该算法相比目前先进的算法有更好的跟踪精度和鲁棒性,且算法平均跟踪速度39帧/s,能够满足实时性的要求.Aiming at the problem that the compression tracking algorithm has the effect of occlusion,fast moving and similar target,we propose the adaptive learning compressive tracking based on Kalman filter（ALCT-KF）.Firstly,the target is located by using the compressive tracking algorithm.Then,the classifier parameters are updated adaptively according to the confidence map of the tracking result.When the target is heavy occlusion,the Kalman filter is used to predict and estimate the position of object.Experimental results show that the proposed algorithm has better tracking accuracy and robustness than current advanced algorithms.And the average tracking speed of ALCT-KF is 39 frames per second,which can achieve real-time performance.

关 键 词：卡尔曼滤波 压缩跟踪 置信图 自适应学习 

分 类 号：TP353[自动化与计算机技术—计算机系统结构]