基于几何参数的步态识别研究

Gait Recognition Based on Gait’s Geometry Parameter

针对人体行走时在空间呈现出的不同几何模式,介绍了一种基于几何参数的步态识别方法,利用Gabor滤波器在不同角度上的滤波特性,对步态图像滤波的基础上,通过扫描二值图像提取人体的步态几何特征:两脚间的宽度、两腿与人体垂直中线的夹角以及两腿交叉点的坐标;步态的分类采用改进后的动态时间扭曲(DTW)方法进行动态模板匹配,基于CASIA步态库的测试表明,该方法的分类准确率达90%。实验结果表明,基于几何参数的步态识别,可以有效地区分不同的运动主体。

Gaits recognition can be used in identifying or intelligent monitor systems, As man walking, their gaits show different geometry patterns, A gait recognition method based on gait＇s geometry parameter is introduced, Firstly a seri^d of gait image is filtered by Gabor filters for its different character in different direction. Then the gait＇s geometry parameters are get by scanning the filtered images. Each image is presented by an abstract point which has four dimensions： step gap, angle fore leg swung refer to middle vertical line of body, angle back leg swung refer to middle vertical line of body and the cross point height two legs crossed, Last reference templates are built and gait patterns are classified based on weighted dynamic time warping matching. Test on CASIA gait Lib, shows that the correction of classification reaches on 90%. Hence, result shows that the gait recognition based on gait＇s geometry parameter can be used to distinguish different moving person effectively.