结合卡尔曼滤波器的改进均值漂移算法

Improved Mean Shift Algorithm Combining with Kalman Filter

在均值漂移框架下,采用帧差法检测运动目标,获取运动信息,同时提取目标参考模型的颜色特征和边缘方向特征,分别计算Bhattacharrya系数,根据Bhattacharrya系数以及前一帧的特征权值进行颜色特征、运动特征和边缘方向特征的自适应加权。此外,根据一定的策略实时更新目标参考模型,以适应运动目标的外观变化。由于结合了三种互补性较强的特征,该均值漂移算法能很好地适应相似的背景颜色干扰、光线变化、目标旋转、突然加速以及尺度变化等复杂视频场景。为了处理目标发生遮挡的情形,将改进的均值漂移算法与卡尔曼滤波器进行有效结合。当目标大部分甚至全部被障碍物遮挡时,仍可以进行稳定的目标跟踪。

Under the framework of mean shift, the frame difference algorithm is applied to obtain the motion infomation. At the same time, the color feature and edge direction feature of the object reference model are extracted, and their Bhattacharrya cofficients are calculated respectively, then according to their Bhattacharrya cofficients and feature weights of the previous frame to achieve adaptatively weighting. Besides, according to one specific strategy to update the object reference model, so it can adapt to the appearance changes of the moving object. By combining three complementary features, the improved mean shift algorithm can adapt to some complex video scenario quite well, such as the background with similar color and light changing, object revolving, sudden accelerating and scale changing. Moreover, in order to handle the circumstance of shelting, kalman filter is integrated into the improved mean shift algorithm effectively. When the object is shelted by some obstacles to a large extent or absolutely shehed even, it can also track the object stablely.