A robust method for camera motion estimation in movies based on optical flow

摘要

Camera motion estimation plays an important role in digital video analysis algorithms such as video indexing and retrieval or automatic movie analysis. Several algorithms have been proposed to solve this problem in MPEG videos. This paper presents an optical flow-based approach for the camera motion estimation in all kinds of digital video formats, especially in movies. It compares the motion vector fields (MVFs) with six predefined templates to determine the type of motion. The MVFs are generated by using high-accuracy optical flow computation. The advantage of the method lies in its robustness to noisy environments such as false motion vectors and object motions. Comprehensive experiments with video clips extracted from well-known feature movies demonstrate the performance of the proposed approach.