Automatic segmentation of moving objects considering single moving camera

摘要

In this paper, we present a new method for detecting and accurately extracting the moving object from a video sequence taken by single moving camera. Extracting high-quality dynamic foreground is a challenging problem due to the coupling of color, motion, and occlusion. Many previous approaches assume that the camera motion is restricted. In our method, we relax these restrictions and present a comprehensive system for accurately computing camera motion, background, and foreground object. Based on the robust motion estimation, we are capable of handling challenging videos where the camera contains unknown motions. The accuracy of estimated camera-motion parameters is increased by direct estimation such that the consistent background is achieved. The foreground separation which includes contextual information decreases the number classification errors, greatly improves the foreground separation even on the object boundary. Experimental results of high quality segmentation of moving objects from challenging videos demonstrate the effectiveness of our method.