关键词: 计算机视觉, 目标匹配, 目标提取, 倾斜度校正, 最小二分之一抽样

Abstract: Target-matching is the research hot-spot in the field of computer vision，while the precision of target-matching is profoundly affected by image segmentation.The segmentation method generally used（two frames difference） always leads the segmentation area wider than the target region，so inaccurate information is introduced.In order to handle this problem，this paper proposes an improved video segmentation algorithm based on the difference of continuous sequential frames combined with Least Half Sampling（LHS） method，which helps greatly to locate the target region accurately.Furthermore，a gradient emendation method is put forward to eliminate the invalidate region caused by sloping road surface.The experimental results illustrate that the proposed algorithm improves the precision and speed of target-matching effectively.

Key words: computer vision, target matching, object extraction, gradient emendation, Least Half Sampling（LHS）