摘要： 针对视频监控中步态识别算法准确度较低的问题,提出一种双边傅里叶校正点估计类Haar特征模板匹配的多镜头步态识别算法。根据视频监控的特点,使用图像点估计重构算法,设计一种无标记类Haar特征模板匹配的步态特征提取算法,进行步态的自动识别与提取。对不同镜头视角下特征运动角度提取差异,基于双边傅里叶级数实现观测角度校正,并依据高维特征空间设计自适应顺序前进浮动选择的搜索算法。在Southampton测试库上的仿真结果表明,该算法步态正确分类率达到96.3%,能有效提高分类识别精度。

关键词: 点估计, Haar特征, 模板匹配, 步态识别, 双边傅里叶级数

Abstract: Aiming at the problem of low accuracy for the gait recognition in video surveillance,the bilateral Fourier calibration point estimation category Haar feature template matching based multiple lens gait recognition algorithm is proposed.According to the characteristics of video monitoring,the point estimation algorithm is used for image reconstruction and a markerless Haar feature template matching based feature extraction algorithm is designed to realize the automatic recognition and extraction of gait feature.According to the angles difference of feature extraction in motion,the bilateral Fourier is used for observation angle correction,and the adaptive sequential forward selection is designed for high dimensional feature extraction.Simulation results in the Southampton test database show that the correct classification rate of the proposed algorithm can reach about 96.3%,and through the observation angle correction,the gait motion angle can reach basically the same in different gait cycles.

Key words: point estimation, Haar feature, template matching, gait recognition, bilateral Fourier series

中图分类号: 

• TP391.41