Moving Objects Detection and Segmentation Based on Background Subtraction and Image Over-Segmentation

摘要

Moving objects detection is a fundamental step in many vision based applications. Background subtraction is the typical method. Many background models have been introduced to deal with different problems. The method based on mixture of Gaussians is a good balance between accuracy and complexity, and is used frequently by many researchers. But it still cannot provide satisfied results in some cases. In this paper, we solve this problem by introducing a post process to the initial results of mixture of Gaussians method. An over-segmentation based on color information is used to segment the input frame into patches. The goal of segmentation is to split each image into regions that are likely to belong to the same object. After moving shadow suppression, the outputs of mixture of Gaussians are combined with the color clustered regions to a module for area confidence measurement. In this way, two major segment errors can be corrected. Finally, by connected component labeling, blobs with too small area are filter out, and the contour of moving objects are extracted. Experimental results show that the proposed approach can significantly enhance segmentation results.