轮廓定位下的人头NSCT特征提取方法

NSCT Feature Extraction Method of Head Under Contour Localization

监控设备目前大量被运用在众多公共场所,其采集的数据信息多数用于对行人的监控。人头目标检测工作也是该领域重要的工作,因为视频数据占用大量的存储空间,为了能减少其空间占存比,市场上绝大多数都会采用低分辨率存储方式,从而导致图像中的信息大量丢失,给人头的精确提取带来极大的影响。针对该类问题,利用人头类椭圆的共性进行人头的快速定位,再利用NSCT特征在轮廓细节上表现出的优越性,对轮廓区域进行NSCT特征提取,最后利用SVM训练器进行分类检测。实验结果表明:该方法能在保证检测实时性的基础上提高在低分辨率环境下的检测效果。

A large number of monitoring equipment are used in many public places. Most of the collected data and information are used for pedestrian monitoring. The detection of head targets is also an important task in this field. Because video data occupies a large amount of storage space,in order to reduce the space occupation survival ratio,the vast majority of the market will use low-resolution storage,which will lead to a large number of image information loss,the precise extraction of the human head had a tremendous impact. In order to solve this kind of problem,this paper uses the commonness of human head ellipses to locate the human head quickly. Based on the advantages ofusing NSCT features in outline details,NSCT feature extraction is performed on the contour regions.Finally,SVM trainers are used to classify detection. Experiments show that this method can enhance the detection effect under low-resolution environment on the basis of ensuring real-time detection.