Occlusion detection using iterative motion estimation and error evaluation

摘要

This paper presents an iterative motion estimation and error evaluation method for efficient occlusion detection. From a pair of reference and target images, the proposed method detects occlusions using four steps: (i) pre-processing for robust motion estimation, (ii) detection of candidates for the occlusion, (iii) one-dimensional motion estimation, and (iv) motion evaluation and update in the target image. The proposed method can be used in various computer vision areas that need an accurate occlusion detection.