基于Kalman预测器的多特征Camshift运动目标跟踪算法  

Moving target tracking based on Kalman predictor and multi-featured Camshift algorithm

作　　者：万中田 冼钟业 胡明宇[2] 陈丽[3] 王雯[2] 肖继来[2] 

机构地区：[1]广东电网有限责任公司佛山供电局,广东佛山528200 [2]武汉大学系统集成与故障诊断实验室,湖北武汉430072 [3]华为技术有限公司,广东深圳518129

出　　处：《武汉大学学报（工学版）》2015年第5期712-716,722,共6页Engineering Journal of Wuhan University

基　　金：国家自然科学基金项目(编号:50677047);南方电网科技项目(编号:K-GX2011-019);湖北省科学条件专项基金项目(编号:2013BEC010)

摘　　要：基于色度特征的Camshift算法在运动目标颜色特征与背景颜色特征差别不大或目标附近有与目标色度相近的物体时,往往会失去跟踪目标或者跟踪目标不准确.据此,提出一种基于Kalman预测器的多特征融合的Camshift运动目标跟踪算法,将色度特征和梯度方向特征结合起来,利用综合直方图实现目标跟踪,并针对运动目标突然加速导致目标跟丢的情况,采用Kalman预测器预测运动目标在下一帧中可能出现的位置,再用Camshift算法搜索目标中心,提高搜索的实时性.实验表明,该改进算法有效地解决了原有算法存在的问题,提高了目标跟踪的速度与精度,满足了实时性要求.Chroma-featured Camshift algorithm may fail to track target accurately and even lose target when ehroma feature is similar between moving target and background or when another object of similar chroma is near the target. Accordingly, moving target tracking based on Kalman predictor and multi-featured Camshhift algorithm is proposed, in which chroma feature and gradient direction feature are combined to a- chieve target tracking using integrated histogram. Also, to prevent the case of losing target caused by sud- den acceleration, Kalman predictor is used to predict possible location of moving target in the next frame; and then Camshift algorithm is used to search for the target center so as to improve real-time performance. Experimental results show that the modified algorithm proposed can improve the speed and accuracy of tar- get tracking, so as to provide more satisfactory real-time performance and thus effectively solve the prob- lems of the original algorithm.

关 键 词：运动目标 跟踪算法 CAMSHIFT KALMAN 多特征 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]