序列图像中彩色目标跟踪的加权颜色分布方法

Visual Target Real-time Tracking Based on the Weighted Color Probability

提出一种基于加权颜色概率分布模型的彩色目标跟踪方法。对目标的颜色直方图进行改造形成目标的颜色概率分布图,同时考虑到了颜色模型的大小及其中像素点的位置对颜色分布的影响,再对颜色分布图进行加权处理成为加权颜色分布图,以此作为目标的颜色模型。与加权直方图相比,加权颜色分布图对区域特征描述更加合理。实验结果表明,本文方法与典型的mean-shift算法相比,提高了目标的定位精度的同时,降低了计算复杂度,而且在快速运动和被大比例遮挡的情况下对彩色目标的实时跟踪有良好的鲁棒性。

This paper presents a real-time color image tracking method. In order to implement an effective and robust tracking task, a novel approach of weighted color target model based on color histogram is discussed, It takes into account the target＇s shape and position of pixels as necessary factors in target model. This makes the description of target＇s feature more reasonable. Furthermore, the starting position of target is found with Kalman filter in every frame, and then mean shift is utilized to track the target position. Experiment result shows, in Contrast with mean shift, the proposed method can locate object more accurately and have lower complication of calculation. The proposed method is more robust, especially for some complex tracking conditions such as occultation and object which is moving fast.