摘要： 针对实时监控识别人形的要求,提出了一种基于矩不变量的分级识别人形“头肩像”技术。首先从视频帧序列中,采用差分的方法分割出活动目标。对活动目标进行预处理后,根据活动目标轮廓的最小外接矩形宽高比、所定义的轮廓描述符和仿射不变矩,依据所选择的分级判别策略,由实验得出的经验数值,判断运动动目标是人形“头肩像”“正面”还是“侧面”。实验结果表明本方法具有快速、适应性强的特点。

关键词: 人形识别, 仿射不变矩, 轮廓跟踪

Abstract: This paper describes a method to detect human based on moment invariants sual surveillance. It uses subtraction from video quences to extract the moving region, after preprocession of the extracted images, caculates the featuers of the the moving region: the ratio between width and heighth of the least circum rectangle ,the description of shape and the affine moment invariants.Based on the graded decision tree and the experiment values, it distinguishes the moving region from human shape and non-huamn shape, the profile and the face.The method is fast and adaptable.

中图分类号: 

• TP391.4