Color Segmentation and Part Model Matching for Non-rigid Objects Tracking

Aiming at improving the performance of non-rigid object tracking in video sequences acquired by a stationary camera, an effective method based on the adaptive color segmentation and object part model was presented. In this work, we modeled background and obtained the foreground blobs with an effective adaptive background updating method based on Gaussian mixture model (GMM), and then the regions in part model were generated online by the color segmentation based on region-growth. The region features and constraints between regions were taken into account and used to perform object tracking effectively and flexibly even under object partial occlusion and deformation. Experimental results with different real-world scenarios demonstrate validity and robust of our solution.