基于卡尔曼滤波的多区域关联运动目标跟踪

针对视频目标跟踪中的遮挡及跟踪漂移问题，提出一种基于卡尔曼滤波的多区域关联运动目标跟踪算法。该算法将目标划分为多个区域并构建无向图，通过卡尔曼滤波预测出各区域中心，再结合灰度直方图匹配及相邻区域的位置关系，计算出各区域观测中心，最后应用卡尔曼滤波修正观测中心实现跟踪。对两区域人体目标跟踪的实验结果表明,与各区域单独采用Mean Shift跟踪算法相比，所提算法在目标遮挡、目标与背景特征相似的情况下，依然具有较好的鲁棒性和实时性。

Moving object tracking with related multi-regions based on Kalman filter

A moving object tracking algorithm with related multi-regions based on Kalman filter was proposed to solve the occlusion and tracking excursion. Through locating multiple regions on the target and constructing undirected graphs, the algorithm calculated the predicted position of the center of each region by using Kalman filter firstly. Then by combining gray histogram match with positional relations of adjacent regions, it calculated the observation center of each region. At last, it realized tracking by revising the observation centers using Kalman filter. The experimental results of human tracking of two regions show that the proposed algorithm has better robustness and real-time performance than the result of tracking each region using Mean Shift algorithm under the occlusion or similar targets and backgrounds.