Current Issue Cover

安国成,陈建军,张凤军,戴国忠(中国科学院软件研究所人机交互实验室,北京 100190;东南大学信息科学与工程学院,南京 210096)

摘 要
Target tracking based on mean shift with improved weights

An Guocheng,Chen Jianjun,Zhang Fengjun,Dai Guozhong()

Aimed to solve the poor tracking accuracy problem of mean shift based on Bhattacharyya similarity, a novel weighting method is proposed based on the enhanced histogram intersection. The paper uses the ratio between the color components of the normalized color probability densities of the candidate model and that of the reference model as the weight coefficient. Results show that the new color weighting coefficient method has good adaptability under fast object moving or color similarity in scene, and the target tracking precision is enhanced effectively. In addition, a template updating method based on auxiliary model mechanism is used to solve the problems of camera vibration, illumination and object changing, therefore the cumulative error problem is also effectively alleviated. Through groups of contrast experiments, the new tracker shows its good adaptive ability and its robustness.