Research On Gait-based Human Identification

In recent years,with the rapid development of information technology,information security has attracted more and more attention,and how to use biometric technology effectively to recognize one's identity has become a research hotspot.Gait recognition is an emerging biometric technology,which aims to identify a person by studying his walking posture.Compared with the traditional finger print,iris,palmprint,face and other biometric technologies,it has the advantages of non-contact,long distance recognition,difficult to hide,low for video definition et al.As an emerging biometric identification technology,gait recognition has important theoretical and broad application prospects.At present,compared with other face and fingerprint recognition,research on gait recognition research is still in its initial stage.Based on gait energy image,this paper makes a deep research and discussion on gait recognition from the following aspects.(1)Gait detection and gait cycles based on video.This paper used background subtraction algorithm to get the gait image sequence,and utilized morphology ideas to reduce noise and holes.Then,we used the periodic changes of aspect ratio of the human body to extract the periodic image sequences.(2)Based on gait energy image,this paper proposed a gait recognition algorithm using Gabor wavelets and collaborative representation.In this algorithm,the Gabor feature of GEI was used as the feature vector,and the gait recognition was performed by the method of collaborative representation.By adding the angle detection strategy,the gait sequence can be identified for visual angle detection to facilitate the same perspective in the database for gait recognition.Experiments on CASIA Dataset B gait database provided by the Chinese Academy of Sciences showed that the algorithm has the advantages of fast running speed and robustness to the cross-view gait recognition.(3)To improve the recognition rate of algorithms using a single feature,a novel gait recognition algorithm was presented in this paper,which was on the basis of feature fusion of dynamic part of GEI and Gabor.Firstly,the gait contour images were extracted through the object detection,binarization,morphological algorithms,and then the GEI was calculated by the gait contour images.Secondly,extracted the information of different angles and directions from dynamic part of GEI by using Gabor wavelets,and reduced the fusion feature by improved KPCA.Finally,the vectors of feature fusion obtained after dimensionality reduction are input to the SVM(support vector machine,SVM)based on multi classification,which is to realize the classification and recognition of gait.Experiments are conducted on the gait database by the Institute of Automation,Chinese Academy of Sciences(CASIA)and the recognition effect is satisfactory.Compared with methods based on single gait feature,the gait recognition rate after the fusion is improved significantly.