Font Size: a A A

Research On Gait Recognition Algorithm Based On Contour Feature Fusion

Posted on:2019-12-20Degree:MasterType:Thesis
Country:ChinaCandidate:P R YuanFull Text:PDF
GTID:2428330566491418Subject:Computer software and theory
Abstract/Summary:PDF Full Text Request
Public security has always been a key issue for people.As a key part of security assurance,identity authentication has become the focus of current research,and accurate identification is its core technology.Benefits from its long-distance,no contact,low video equipment requirements,and low camouflage,gait recognition has become a research direction for a large number of researchers.In the related research of gait recognition,the method based on contour features has become a research focus in recent years because of its low computational complexity and low requirement on image quality.This paper focuses on the fusion of multiple contour features on,the basis of contour features.First of all,complete target is extracted from video image.The background subtraction method is used to extract the target contour from the image;and the image is binarized.And the morphological method is used for denoising,so as to obtain a target image.Considering that gait is a kind of cyclical exercise behavior.Therefore,this paper uses the periodic changes of the width of the contour as the basis to detect the gait cycle of the target.The target image is standardized to reduce the effect of different camera shooting distances on recognition to a certain extent.Secondly,the contour of the target image contains a large amount of gait information,and there are many features that can be used for recognition.Aiming at the problem of incomplete gait information for single feature,this paper extracts three kinds of features,such as key distance feature,contour width feature and wavelet feature,and uses the nearest neighbor classifier to identify and verify.The recognition rate of 70%,73.33%and 91.67%is achieved for three features at 90° view.Finally,the three features are fused to recognize,and the recognition rate reaches 98.33%.At the same time,there are more than 90%recognition rates under other views.Finally,aiming at the problem of containing less dynamic information of contour when extracting features from outer contour of the target,this paper proposes a gait recognition algorithm combining gait energy image and contour features.First,reducing the dimensionality and calculating the mean difference of gait energy image.Then the wavelet features are merged with it.Finally,the nearest neighbor classifier is used to identify and verify.The fused features not only have good performance at multiple views,but also have robustness to different walking conditions.Experimental results show that the recognition effect of multiple contour features fusion is obviously better than single feature,and the gait feature recognition effect supplemented with internal information is also better than a single contour feature.At the same time,it has certain robustness to factors such as angle,clothing,and backpack.
Keywords/Search Tags:gait recognition, contour feature, gait energy image, feature fusion, nearest neighbor classification
PDF Full Text Request
Related items