一种抗遮挡的运动目标跟踪算法

提出了一种基于彩色特征的抗遮挡目标跟踪算法。利用mean shift递推寻找当前帧目标的位置,并通过Kalman滤波估计目标状态。选用对目标部分遮挡具有鲁棒性的加权量化彩色直方图作为目标特征的概率分布,用Bhattacharyya系数作为特征相似性度量。提出一种目标遮挡因子,作为目标被遮挡程度的判断根据。当目标严重遮挡后,观测位置不再满足Kalman滤波的条件,采用目标状态量外推取代Kalman状态更新来预测目标当前的位置。实验结果表明,此方法对于部分遮挡以及全遮挡有较好的鲁棒性。

Anti-occlusion arithmetic for moving object tracking

A new method with occlusion for color-based object tracking is presented. The proposed technique employs mean shift iterations to derive the object candidate which is the most similar to a given object model, then uses Kalman filter to estimate the real states of the object. A color-based histogram with different weights that is robust for partial occlusion is selected as the target feature. The similarity between the target model and the candidates is expressed by a metric on the Bhattacharyya coefficient. An occlusion coefficient is proposed. When the object is occluded seriously, the observation cannot be used for updating by Kalman filter, the former state of the object is regarded as the current state. The simulation experiments occlusion. show that the tracking is robust to partial and serious