随着智能化程度的逐步深入，公共场合的安全形势日益严峻，如何确保公共安全成为人们关心和瞩目的焦点，而有效的身份识别技术则是确保公共安全的关键。生物特征以其安全、稳定、可靠等特点广泛应用于智能监控领域，较之于人脸、指纹等，步态以其非侵犯、远距离、难以隐藏等优势受到了大批研究者的关注。近年来，多视角下的步态识别问题一直是步态识别研究的一个热点，因此本文在多视角的背景下，重点研究了步态特征的提取以及视角归一化问题。本文完成的主要研究工作如下： 首先，采用背景差分法完成目标的提取。针对目标图像中含有噪声及不连通等问题，对其采用形态学方法进行去噪，并进行连通区域分析来获得较完整的二值化步态图像。由于步态是一个周期性的运动，一个周期内的步态变化更能反映人体的运动特性，因此本文根据步态轮廓宽度变化来检测步态周期，并对步态图像进行标准化处理。 其次，人体的步态图像序列不仅含有静态的步态信息，同时相邻步态间的变化也隐含了丰富的动态信息，而常用的步态能量图和步态帧差能量图只考虑了其中的静态信息和部分动态信息，因此本文将刻画不确定性的熵引入到步态帧差能量图中，提出采用步态帧差熵图刻画步态特征，再采用最近邻分类法完成分类识别。 最后，针对多视角步态识别过程复杂、计算量大等问题，本文提出基于低秩优化的视角归一化方法进行步态识别。在步态特征图像的基础上，将任意视角下的步态特征图像采用秩优化的方法归一化到秩最小的视角，再采用最近邻分类法进行跨视角下的识别验证。 实验结果表明，本文提出的步态帧差熵图的识别率要高于步态能量图和步态帧差能量图，而基于步态帧差熵图的视角归一化步态识别方法也在一定程度上提高了跨视角下的识别率。

Along with the deepening of intelligence, the security in public places is increasingly serious, and how to ensure the public security has become the focus of public concern and attention, while the effective identification technology is the key to ensure public security. Biological characteristics are widely used in intelligent monitoring areas based on its advantages of security, stability, reliability, and so on. Compared with face, fingerprints et al, gait is favored by a large number of researchers because of its superiority of non-invasive, non-contact and difficult to hide. In recent years, Multi-view gait recognition has always been a hotspot in gait recognition. So under the background of multi-view gait recognition, this paper mainly focuses on the extraction of gait features and the normalization of perspective. The main work done in this paper is as follows: Firstly, extract object adopting background difference method. In order to solve the problem of noise and disconnection in the object image, the morphological method is used to denoise, and the connected region analysis is carried out to obtain a more complete binarized gait image. Gait is a periodic motion, and gait changes in a cycle can better reflect the movement characteristic of a human. Therefore, this paper detects gait cycle according to the change of the gait silhouette width, and then normalizes the gait image. Secondly, the gait image sequence of human not only contains the static gait information, but also the change of the adjacent gait implies the rich dynamic information. While the commonly used gait energy image and gait frame difference energy image only consider the static information and some dynamic information, so this paper introduces entropy describing uncertainty to gait frame difference energy image, proposes gait frame difference entropy image to describe gait feature, and then uses the nearest neighbor classification method to complete recognition. Lastly, In order to solve the problem of complex process and large computation in the multi-view gait recognition, this paper proposes a method of view normalization based on low rank optimization to recognize gait. On the basis of gait feature image, the gait feature image at any view are normalized to the view with the minimum rank using the method of low rank optimization, and then the nearest neighbor classification method is adopted to recognize gait at cross-view. The experimental results show that the recognition rate of gait frame difference entropy image proposed in this paper is higher than gait energy image and gait frame difference energy image, and the view normalization gait recognition method based on gait frame difference entropy image improves the recognition rate at cross-view to some extent.