基于改进的颜色特征和直方图更新的Mean-Shift算法

摘要

In this paper, a target tracking algorithm of mean-shift based on improved color feature and histogram updating is proposed to enhance efficiency of intelligent video surveillance in complicated scene. The variety of color around a pixel is taken as the feature of the pixel to improve the single color' s capability of expressing the target. The probability density of image is easily forecasted based on the histogram to enhance the efficiency of mean-shift. The histogram of target is updated based on matching degree between the target and the result of mean-shift. The experimental data indicates that the improved color feature can avoid interference of similar pixel of background and that the updating of histogram can solve the problem of the change of target's size. This algorithm can improve mean-shift' s validity of the real-time target tracking in the complicated scene.%为了提高复杂环境下智能视频监控对运动目标跟踪的可靠性,提出一种基于改进的颜色特征和直方图更新的Mean-Shift跟踪算法.该算法利用像素点局部颜色变化作为跟踪目标像素点特征,弥补单一颜色值特征对目标表征的不足；并根据Mean-Shift跟踪结果和目标匹配程度对目标直方图进行更新,保证在目标姿态和大小发生变化时目标特征直方图的有效性.实验结果表明:与传统Mean-Shift算法相比,改进的颜色特征减小了相似背景像素对运动目标的干扰,目标直方图的更新提高了Mean-Shift算法对目标姿态和大小改变的鲁棒性.算法提高了基于颜色特征的Mean-Shift算法对复杂环境中运动目标进行实时跟踪的可靠性.