正面视角的步态识别

摘要

现有的步态识别算法研究几乎全都是基于侧面步态的.提出一种基于正面视角的步态识别方法.首先归一化RGB颜色空间被用来检测和去除阴影,并用背景减除法提取二值化人体轮廓.提出一种专门适用于正面步态的周期检测方法,提取周期关键帧后跟踪轮廓线,并用改进的等角度采样法进行采样以减少计算量.简单高效的傅里叶描述子被用来提取特征向量,进行数据降维后构造步态模板.用最近邻和最近邻标本分类器分别进行分类.在CASIA数据库上的实验表明,该算法不仅具有较低的计算量而且表现出较好的识别性能.%Existing gait recognition methods are usually based on side view sequences; however, in this paper a new method for front-view gait recognition was proposed. First, a method of normalized RCB color space was used for detection and shadow removal and the binary silhouettes were extracted by background subtraction. Then, a special method for cyclic gait analysis based on a front view was performed to extract cyclic key frames. Next, body contours were tracked, and an improved equal-angle sampling method was applied to reduce the number of computations. Next, the feature vectors were extracted efficiently. Fourier descriptors as data dimension were reduced, allowing the gait template vectors to be constructed. Finally, recognition was achieved separately by the nearest neighbor classifier NN and the nearest neighbor classifier with respect to the template ENN. Experimental results show that the proposed approach is not only efficient in computing, but also has an encouraging recognition performance in the CASIA database.