一种新的AdaBoost视频跟踪算法.pdf

第27卷第5期 Vol. 27 No. 5 控制与决策 Control and Decision 2012年5月 May 2012 一种新的AdaBoost视频跟踪算法文章编号:1001-0920 (2012) 05-0681-05 徐建军 1,张蓉 2,毕笃彦 1,孙路 3 (,西安710038;, 北京100195;,北京100085) 摘 要:针对复杂场景中运动目标较难定位的问题,;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 ,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. 关键词:目标跟踪;AdaBoost;Gabor变换;均值漂移中图分类号:TP391 文献标识码:A An new AdaBoost video tracking algorithm XU Jian-jun 1,ZHANG Rong 2,BI Du-yan 1,SUN Lu 3 (1. School of Engineering,Air Force Engineering University,Xi’an 710038,China;2. 95961 Air Force Army, Beijing 100195,China;3. Radar & EW Institute,Beijing 100085,:XU Jian-jun,E-mail: xujianjun?y@) Abstract:To solve the problem of moving objects located dif?cultly plex background, a tracking algorithm based on AdaBoost is proposed. Firstly, an ensemble of weak classi?ers is trained online to distinguish between the object and the background. Then, the ensemble of weak classi?ers bined into a strong classi?er by using AdaBoost, and strong classi?er is used to label pixels in the next frame, and a con?dence 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 illu