Object Tracking Based on Partition and Recombination
This paper presents a new approach for solving the scale selection problem in object tracking using Mean Shift algorithm. Although Mean Shift algorithm is an effective method for object tracking, scale selection in Mean Shift is still an important and difficult task. Simply speaking, the scale selection is to find a kernel window in suitable size for the representation of the moving object in the process of tracking. Choosing the scale too large or too small will both cause mistake. In our work, we propose an effective method based on object partition and recombination. This proposed method partitions the detected object into two sub parts and tracks each sub part individually, then adjusts the windows size through segmentation, finally, our approach combines above steps into a new window. The experimental results show the proposed method is superior to the traditional methods in scale selection of Mean Shift algorithm.
Mean Shift scale selection object tracking
Yinghuan Shi Yang Gao Liangdong Shi Xiaowen Guo
State Key Laboratory for Novel Software Technology Nanjing University,China
国际会议
2008高等智能国际会议(2008 International Conference on Advanced Intelligence)
北京
英文
2008-10-18（万方平台首次上网日期，不代表论文的发表时间）