Gait recognition, as a new kind of biometrics recognition method, refers to automaticidentification of an individual based on his/her walking style.Compared with facerecognition, fingerprint recognition and so on, it has prominent characteristics of longdistance, non-contact, difficult to camouflage and difficult to hid, and it is the mostpotential biometric recognition in the field of remote identification. In addition, it has abroad application prospect at security system, human ID management, and digitalsurveillance, and has become a research hotspot in recent years.A complete gait recognition process usually consists of four stages: target detection,cycle detection, features extraction and classification recognition. Existing gait recognitionalgorithms are almost based on side view. It’s rare to find researches on other walkingdirections, especially the front-view direction, so gait recognition of considering walkingdirection has become a challenging problem. As a result, this paper focuses on front-viewgait recognition algorithms.First, the traditional target detection methods are used for fronted target detection, thehuman body binary image often appears large hole, or in the body of the lower limbs areaappears large shadow and so on, this paper puts forward an improved method ofbackground difference by the threshold binarization at obove and lower part of humanbody, and then the complete binarization of the human body contour can be got afterimage post-processing.For fronted cycle detection, first of all, using hip ideas can find the optimal point ofdividing human left and right leg (the point P), then three new phase detection method isproposed according to the point P: Using the pixel difference between the human foots forcycle detection; Using the distance between human foot bottom for cycle detection; Usingthe tangent value of human lower limb angle for cycle detection.Then, in view of the present positive gait recognition rate is low and gait feature issingle, three new algorithms of gait recognition are proposed based on feature fusion inthe paper: Based on feature fusion recognition algorithm of unified Hu moment and gait cycle; Based on feature fusion recognition algorithm of unified Hu moment and humanlower limb angles; Based on feature fusion recognition algorithm of unified Hu moment,gait cycle and human lower limb angles. Then comparing the three algorithms, the resultsshow that the above three characteristics of the fusion recognition rate reach the highest,so as to solve the low rate recognition problem which is caused by the one single feature.The support vector machine (SVM) is used for classification. The recognitionexperiments of this paper are all maked in the Chinese Academy of Sciences (CASIA) gaitdatabase, the experimental results show that the methods proposed in this paper achieve ahigher recognition rate, and all the methods are feasibil. |