线性插值框架下矩阵步态识别的性能分析

Performance analysis of matrix gait recognition under linear interpolation framework

针对现有的步态周期检测方法检测效果不佳以及行走速度变化对步态识别性能有很大影响的问题,提出的基于矩的步态周期检测方法中,Zernike矩需要人体居中、尺度归一的前期预处理过程,而伪Zernike矩具有能描述运动图像的特点,它可以避免人体居中、尺度归一等处理,以便直接测试步态的周期性.根据行走时的两帧之间的特征取决于前一帧和后一帧的特征,提出了基于线性插值的矩阵步态识别算法框架,并且将投影特征、Hough变换特征、Trace变换特征和Fan-Beam映射特征应用在CASIA(B)步态库上,验证了框架的有效性,为解决步态识别问题带来新的方法与思路.这种基于线性插值的矩阵步态识别特征本质上是一种权值不同的能量形式.

The existing gait period detection methods are not ideal and the performance of gait recognition is signifi-cantly influenced by walking speed .Several novel gait period detection methods based on moments are proposed in this paper .The Zernike moment requires preprocessing including the assurance that the image of the human body is proportioned normally and is centered properly; the pseudo-Zernike moment may directly describe the motion im-age, and it may avoid the need for such processing of making the image of the human body centered and sized nor -mally, so as to directly detect gait periodicity .As the features of one frame are only decided by those of the prior and the rear frames in walking , a framework for a matrix gait recognition algorithm based on linear interpolation is proposed.Subsequently, the projection features, Hough transform feature, Trace transform feature and Fan-Beam mapping feature are applied to the CASIA （ B） gait database to prove the validity of the gait recognition framework . This brings new methods and understanding for solving gait recognition problems .This matrix gait recognition feature based on linear interpolation is essentially an energy form with different weighted values .