Common Methods Of The Gait Recognition

Gait recognition as a new application in the computer vision, get more and more technical personnel's attention.In recent years the States are focus on terrorist attacks,researchers need to find a way to execute video surveillance from distance,the gait recognition has the feature of non-contact,non-invasive and difficult to hide,which give a method to technical personnel for video surveillance.Gait is defined as"a manner of walking"in the Webster Collgiate Dictionary.We extend our definition of gait to include both the appearance and the dynamics of human walking motion,Johansson had shown in the 1970's that observers could recognize walking subjects familiar to them by just watching video sequences of lights affixed to joints of the walker.So,in theory,joint angles are sufficient for recognition of people by their gait.However,recovering joint angles from a video of walking person is an unsolved problem.So,we have included appearance as part of our gait recognition features.This article is a reviews,which mainly related to the various implementation stages of gait recognition.including background modeling,object detection,object of the pretreatment,cycle detection,feature extraction,data processing,classification and identification.The first chapter of this article mainly introduce the current status of gait recognition,which strating from the biometric identification,give the chart,and the feature of the gait recognition:non-contact,non-invasive,hard to hide.Finally,the overall framework of the article is given.The second chapter of this article give introduction of the preprocessing,the current gait recognition's algorithms are generally based on a binary image library of the gait,so the preprocessing is related rarely.The preprocessing contains the background modeling,object extraction and targets'preprocessing:background modeling contains Kernel Density Estimation and least square mean;object extraction contains the largest mean variance and indirect difference method;morphological processing.In the area of Kernel Density Estimation,we give the calculation process and get the answer that Kernel Density Estimate models can handle multi-background modeling and have a good robustness to the change of the background,at the same time,we show the reduce of the calculation using the Diversity sampling scheme.The third chapter of this article is the specific process of gait recognition,including the cycle of detection,feature extraction,data processing,classification and recognition.Specifically,cycle of detection contains height and width based on silhouette,the number of pixels with the lower body;feature extraction contains the model-based and the non-model based,the model-based includes elliptical model and the three-dimensional model.the non-model includes temporal correlation matching method,key-frame method,contour points distance,gait energy image and optical flow features;data process is mainly the extraction of the data for the principal component;classification and recognition,including support vector machine and K-nearest neighbor.In the area of spatio-temporal pattern,we know that this is robust to the noise,and have the little calculated amount.In the area of the gait energy diagram,we know that this method has the little calculated amount although this method has lost some information on the gait.In the area of PCA,we get the main idear about the PCA,we transform one data space to another data space which reduce the dimension of the original data space,and emphasize that make the classification and recognition in one data space.In the area of the SVM,we give the tools to use the SVM,and have the process of using the libsvm. The last chapter of this article give some advice of mine:we give the summary from recognition rate,noise,performance evaluation and have a prospect from 3-D modeling and multi-camera,data fusion,the large gait database.