基于卡尔曼和图像信息量的MeanShift改进跟踪算法  

Improved Mean Shift Tracking Algorithm Based on Kalman and Image Information Measure Method

作　　者：王梦斐[1] 王沛[1] 马燕[1] 张倩[1] 

机构地区：[1]上海师范大学信息与机电工程学院,上海200234

出　　处：《电视技术》2015年第5期41-44,86,共5页Video Engineering

基　　金：国家自然科学基金项目(61373004)

摘　　要：Mean Shift算法在目标实时跟踪领域取得了广泛的应用,但是对于速度过快或尺度变化大的目标跟踪存在较大的缺陷。提出了一种基于Mean Shift和Kalman预测带宽的自适应跟踪算法。该算法提出以Kalman预测目标在下帧中的中心位置作为Mean Shift迭代初始位置;同时引入图像信息量度量方法以适应目标的尺度变化。实验结果表明,改进的跟踪算法能很好地跟踪尺度变化的目标,跟踪效果很好。Mean Shift algorithm is widely used in high real-time field of target tracking. However, it cannot adapt to the size change of the target or locate accurately when the target is moving fast. In this article, a bandwidth-adaptive tracking algorithm based on Mean Shift and Kalman prediction is proposed. The Kalman filter is used in the algorithm to predict the positions of fast moving objects in the successive frame, which are as the initial positions for Mean Shift tracking. Also, the information measure of multi-scale image in scale space is used to differentiate the scale. The experimental results indicate that the improved algorithm can adapt to the size change of the target and the tracking result is improved.

关 键 词：视觉跟踪 均值漂移 信息度量 卡尔曼 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]