针对在复杂背景下跟踪运动目标的要求, 建立了目标的显著性直方图模型, 提出了改进的连续自适应均值漂移(Camshift)跟踪方法。通过比较目标区与背景区的色调差异, 计算目标不同色调等级的显著性值; 基于加权的方式强化显著性色调在目标识别过程中的作用, 弱化非显著性色调的作用, 从而抑制背景区对目标识别的干扰。利用加权直方图模型反向投影, 建立了跟踪图像的概率投影图, 利用均值漂移方法完成目标跟踪任务。将该方法分别应用于标准测试库视频图像的跟踪以及实际运动目标的跟踪实验中并与传统方法进行了比较。结果显示, 该方法能够利用显著性色调很好地将目标从背景中区分出来, 在计算量增加不多、且满足电视跟踪系统实时性要求的情况下, 提高了目标识别的准确性和稳定性, 目标定位的最大偏差与被跟踪目标区的尺寸比小于25%, 能够确保被跟踪目标不丢失。

According to the target tracking requirements in complex backgrounds, an improved Continuously Adaptive Meanshift(Camshift) tracking method was proposed by modeling a saliency histogram of the target. The saliency values of different hues in the target area were calculated by comparing the difference between the target and the background area. The weighted histogram was used to strengthen the roles of the saliency hues and weaken the roles of the non-saliency hues, by which the interference from the background was restrained. By using the back projection of the weighted histogram, the probability projection image of the tracking image was obtained by the back projection, then the target tracking task was completed by mean shift method. The proposed method was applied to an actual target in tracking experiments and that in the video of the standard test libraries and obtained results were compared with that of traditional methods. The simulation results show that the target is easily recognized from the background by the saliency hues, and the accuracy and the stability of the target recognition are improved with satisfied real time ability and without too much computation cost. The ratio of the max deviation to the size of the target is less than 25%, which ensures the target not to be lost.