结合卡尔曼滤波和Mean Shift的抗遮挡跟踪算法

Anti-Occlusion Tracking Algorithm Combined Kalman Filter and Mean Shift

针对卡尔曼滤波和Mean Shift算法结合后对严重遮挡和遮挡后复出失效且实时性差的问题,提出一种基于卡尔曼滤波和Mean Shift动态结合的改进算法.通过在算法中加入Bhattacharyya系数进行遮挡程度判断,并根据遮挡系数的阈值选择使用卡尔曼滤波或线性预测法更新Mean Shift迭代起点.实验结果表明,该方法能成功实现大范围连续遮挡和目标复出情况下红外目标的跟踪,并且迭代次数和跟踪时间分别减少了9.68%和17.58%,提高了跟踪的鲁棒性和实时性.

To solve the problem of significant occlusion and failure when reappearing in combining Kalman filter and Mean Shift, a new improved method which is based on Kalman filter and Mean Shift was proposed. In the algorithm, first, the parameter of Bhattacharyya is used to scale the degree of occlusion, then Kalman filter or linear prediction was chosen to update the searching loop point of Mean Shift according to the Bhattacharyya parameter. The experiment results indicate that the searching and tracking time can be reduced down 9. 68% and 17.58%. A continuous and stable tracking results can be obtained in the situation of significant occlusion and re-appearance.