An improved mean shift algorithm for moving object tracking
Xiaoping Chen, Shengsheng Yu, Zhilong Ma · 2008
Traditional mean shift algorithm requires a symmetrical kernel, such as a circle or an ellipse, and assumes the kernel represents the object shape. Because the symmetrical kernel always contains some background regions, the performance of moving object tracking is dramatically affected when background is complex and changes greatly. To address above issue, this paper proposes an improved mean shift algorithm, which first performs image segmentation to obtain object shape from the selected region, then uses the object shape to construct a level set asymmetric kernel for mean shift. Therefore, the impact of changes in background is greatly reduced. Experimental results show that the algorithm presented in this paper is more effective and robust than traditional mean shift algorithm.