Analysis Of Human Gait Based On ViBe Algorithm At Specific Viewing Angles

With the advancement of science and technology and the progress of the times,human information security is facing great challenges,and authentication technology has become a research hotspot.Biometrics technology is an automatic identification technology based on the individual's unique physiological or behavioral characteristics.It is a more reliable and convenient popular identification technology than the traditional identity authentication technology,because the biometric features are not easily transferred or stolen as the identity cards.From the view of visual surveillance,gait recognition has the advantages of easy acquisition,long-distance,non-touch,difficult camouflage and non-invasion,so gait is the most potential biological feature under long-distance conditions.This paper mainly focuses on foreground detection,feature extraction,classification and recognition.The main research work done in this paper includes:1.In the foreground detection.Based on ViBe(Visual Background extractor,ViBe)algorithm,an improved method of suppressing neighborhood update is proposed.And for different viewing angles,the improved ViBe algorithm results are used to fuse with binary images.Experiments show that ghosts and background noise can be effectively removed.2.For the problem of indoor scene shadows,an improved shadow removal algorithm is proposed,which is normalized according to the proportion of each color component value R,G and B.Increasing the rate of change of brightness as a shadow determination condition,and the shadow is removed for two times for the whole target and the lower quadrant of the whole target.Experiments show that the removal effect of shadows is better.3.Feature extraction and classification recognition.Human gait features are analyzed such as periodicity etc.from the database image sequences.Experiments have shown that the periodic features are more obvious if the side viewing gait adopts the length-width ratio of contour-rectangle and the front viewing gait adopts the ratio of pixels in the lower quarter of the human body.Then adaptive normalization is used to remove redundant data,Finally,the gait energy image and Hu moments are used to extract features,then k-Nearest Neighbor(kNN)classifier is used to classify and recognize.Experiments show that the recognition rate of features extracted from gait energy image is higher.The paper selects 0° and 90° image sequences in CASIA-Dataset-B(normal walking)for gait recognition.Experiments show that the algorithm could obtain better results.