Human gait recognition: A silhouette based approach

Our results illustrated that feature extracted from the averaged silhouettes which in them, the lower part of the body is eliminated are more suitable rather than those extracted from the complete averaged silhouettes.
Nội dung trích xuất từ tài liệu:
Human gait recognition: A silhouette based approachJournal of Automation and Control Engineering, Vol. 1, No. 2, June 2013Human Gait Recognition: A Silhouette BasedApproachNegin K. HosseiniFaculty of Information Science & Technology, Universiti Kebangsaan Malaysia, Selangor, MalaysiaEmail: negin62_k@yahoo.comMd Jan NordinCenter for Artificial Intelligence Technology, Faculty of Information Science & Technology, Universiti KebangsaanMalaysia, Selangor, MalaysiaEmail: jan@ftsm.ukm.myAbstract—Human gait has become an important biometricin recent years. A silhouette based method is suggested inthis paper, to recognize human in video by their gait. Weused averaged silhouette to represent the gait cycle.Principal Component Analysis has been used to reduce thedimensionality of the features. We applied Euclideandistance to measure the similarity of the averagedsilhouettes. We implemented the algorithm on the TUMIITKGP Gait Database which has been introduced recently.Although this method is sensitive to the appearance of thesubject, it has low computational cost and it is simple. Weimplemented two experiments on the achieved averagedsilhouettes. Our results illustrated that feature extractedfrom the averaged silhouettes which in them, the lower partof the body is eliminated are more suitable rather than thoseextracted from the complete averaged silhouettes.Adelson in 1994, their methodology was extractingspatiotemporal features from the subject’s gait forrecognition [4]. Afterwards, several studies implementeddifferent gait recognition algorithms [5]-[7].Gait recognition methods are generally divided intotwo different categories: model-based and appearancebased. Model free gait recognition methods or appearancebased methods work directly on the gait sequences. Theydon’t consider a model for the human body to rebuildhuman walking steps. They have the advantage of lowcomputational cost in compare with model-basedapproaches and they also have the disadvantage ofsensitivity to cloth and appearance changing. There areseveral appearance based attempts in order to solve gaitrecognition problem [8]-[10].Model-based approaches are those approaches whichbuild a human body model and the extracted features ofgait sequences will be fitted to that model. Model basedapproaches almost are not sensitive to the individual’sappearance and clothing. On the other hand, model basedapproaches have high computational cost. Niyogi andAdelson [4] suggested the first model-based gaitrecognition approach by modeling human body into 5sticks (2 sticks per legs, 1 stick for the body). Afterwards,several model-based approaches have been suggested byresearchers [11]-[13].Index Terms—gait recognition, averaged silhouette,principal component analysis, euclidean distanceI.INTRODUCTIONThe study of approaches to identify a human beingbased on physical or behavioral traits such as face,fingerprint, ear, voice, gait, iris, signature, and handgeometry is called biometrics. Each biometric has itsrelative benefits in various operational situations.Therefore, it is obvious that no single biometrics isexpected to effectively fulfill all our concerns (e.g.,accuracy, practicality, cost). Several human recognitionapproaches, such as fingerprints, face or iris biometrics,generally require a cooperative subject, or physicalcontact. These approaches can’t be applied for noncooperative subjects or in surveillance scenarios thatidentifying in distance is required. Gait recognition that isbased on the way human walks is a biometric that iswithout the above-mentioned disadvantages [1].Biomechanics [2] and psychophysical [3] studiesillustrated that it is possible to achieve an almost uniquesignature from each individual’s gait. First attempt of gaitrecognition in computer science was done by Niyogi andII.Human recognition based on gait is generally done byextracting the silhouette of the walking subject andanalyzing it during walking. This paper proposes anappearance based recognition method applying theaveraged silhouette of a subject during a gait cycle. Ourproposed gait recognition approach consists of three basicstages: Human detection and Tracking, FeatureExtraction and Training and Recognition. Fig. 1illustrates an overview of the implemented method andeach stage is described in details by the followingsections.Manuscript received September 15, 2012; revised December 22,2012.©2013 Engineering and Technology Publishingdoi: 10.12720/joace.1.2.103-105PROPOSED METHODOLOGY103Journal of Automation and Control Engineering, Vol. 1, No. 2, June 2013C. Averaged Silhouettes ComputationThe gait representation method that applied in thispaper is Averaged silhouettes [15]. Given all silhouettesin a complete gait cycle Gc  Gc(1), Gc(2),, Gc( N ) ,which N is the number of binary silhouettes in a gaitcycle. In order to achieve a set of averaged silhouettes,Avs  Avs (1), Avs (2), Avs (i) , we need to computeaverage of silhouettes in a gait cycle. For each subject,the averaged silhouette ( Avs (i )) is achieved byAvs (i ) 1 N Gc(l )N l 1(1)An example of the averaged silhouettes achieved inthis study, is illustrated in Fig. 3.D. Feature ExtractionEigenspace transformation which is based on PrincipalComponent Analysis (PCA) [16] is applied to theaveraged silhouettes extracted from previous step. Thisstep is implemented to reduce the dimensionality of thefeature space. Let Avs1, Avs2 ,, Avsm be a set ofm averaged silhouettes. The largest eigenvectors of thematrixFigure 1. An overview of the implemented method.A. Human Detection and TrackingGenerally, the first step in a gait recognition system isdividing video frames into background and foreground.This step’s goal is achieving the binary silhouette of thewalking subject. Since, we applied black and white gaitsequences fr ...