一种新的AdaBoost视频跟踪算法  

An new AdaBoost video tracking algorithm

作　　者：徐建军[1] 张蓉 毕笃彦[1] 孙路 

机构地区：[1]空军工程大学工程学院 [2]空军95961部队 [3]雷达与电子对抗研究所

出　　处：《控制与决策》2012年第5期681-685,共5页Control and Decision

基　　金：国防重点实验室基金项目(9140c610301080c6106)

摘　　要：针对复杂场景中运动目标较难定位的问题,提出一种结合纹理和颜色特征的AdaBoost目标跟踪算法.首先在线训练一个弱分类器的集合区分目标和背景;然后,通过AdaBoost将集合中的各弱分类器组合成一个强分类器,用于标定下一帧中各像素的类别属性,并生成置信图;最后,在置信图中用Mean Shift算法定位目标的中心.实验结果表明,该算法在光照变化、目标自身发生形变和遮挡的情况下,能准确地对目标进行跟踪.To solve the problem of moving objects located difficultly under complex background,a tracking algorithm based on AdaBoost is proposed.Firstly,an ensemble of weak classifiers is trained online to distinguish between the object and the background.Then,the ensemble of weak classifiers is combined into a strong classifier by using AdaBoost,and strong classifier is used to label pixels in the next frame,and a confidence map is given.Finally,the new position of the object is found by using mean shift algorithm.Experimental results show that this algorithm is robust and can track the object accurately under the conditions of illumination variation,shape change of object and partial occlusion.

关 键 词：目标跟踪 ADABOOST GABOR变换 均值漂移 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]