融合两层卡尔曼滤波和Mean Shift的自适应目标跟踪算法

Adaptive Object Tracking Algorithm Using Mean Shift Combined with Two Layers of Kalman Filter

为了改进Mean Shift算法及其与卡尔曼滤波融合跟踪算法的性能,提出了融合两层卡尔曼滤波和Mean Shift的自适应目标跟踪算法。首先通过运动学方程建立第一层的数学模型;然后利用巴氏系数、滤波器噪声与跟踪结果之间的关系,自适应地调整跟踪结果,得到目标的位置;最后对目标核函数直方图中的每个非零元素进行第二层滤波,通过动态变化的滤波残差和巴氏系数,实时调整更新滤波器中的各项参数,得到滤波后的目标模板。实验表明,该文算法与Mean Shift算法和单层卡尔曼滤波算法相比,在目标遮挡、光照变化和复杂环境下的跟踪效果更好。

An adaptive tracking algorithm that combined Mean Shift with two layers of Kalman filter was proposed in order to improve the performance of Mean Shift and Mean Shift with Kalman filter, First, we established mathematical model of first layer through kinematics equation; then, we used the relationship among Bhattacharyya coefficients, filter noise and tracking results to adjust self-adaptively tracking results to get target position. At last, in the second layer, we filtered all nonzero elements of the object kernel histogram through dynamic filter residual and Bhattacharyya coefficient, got filtered target template by adjusting and updating parameters of the filter in real time. The experimental results showed that compared with Mean Shift and Mean Shift with single layer of Kalman filter, we got better tracking results under the occlusion, changing illumination and complex environment.