基于矩不变量的二维图象序列动态特征抽取  

Dynamic Feature Extraction in a Two-dimansional Image Seguence Pased on Moment Invariants

作　　者：胡大可[1] 蒋志新[1] 

机构地区：[1]浙江大学科仪系

出　　处：《信号处理》1990年第1期9-16,共8页Journal of Signal Processing

摘　　要：本文提出用图象序列分析和特征抽取的方法来自动识别人体关节点,建立起基于矩不变量的二维图象序列动态特征抽取的数学模型,并从信息论的观点出发给予了分析,详细论述了人体关节点匹配中的特征表、搜索区域的动态建立等概念以及两步法模板匹配原则。在对跳高运动员的技术影片分析中,详细讨论了序列特征的“表现”,达到了用计算机自动识别人体关节点的目标。By means of image seguence analysis and feature extraction, in this paper, the method of automatic recognition of joint points of the human body has been put forward. The consruction of mathematical model for dynamic feature extraction in a two-dimensional image seguence is based on moment invariants, which has been analysed from the point of view of information theory. In the algorithm for matching of joint points of the human body, the concepts of a feature table and dynamic construction of a Searched aera and the princile of two-stage templating matching have fully been stated. In the analysis of a high-jumper's technigue film, the behavior of features in sequences has been discussed in detail. The goal of automatic rcognition of Joint points by computors has been attained.

关 键 词：图象识别 运动员 关节 计算机 跳高 

分 类 号：G804.4[文化科学—运动人体科学]