Research And Implementation On Gait-based Identification

With the increasing requirements for security, video surveillance inpublic places and facilities has become omnipresent. Humanidentification from the information provided by video surveillance isbecoming an urgent requirement. Gait recognition, aiming to identifyindividuals by the way they walk, is a relatively new biometricidentification technology. Compared with other biometric features, gaithas its unique advantages. It does not require users’ interaction and it isno-invasive. Also it is difficult to conceal or disguise. Furthermore, gaitcan be effective for recognition at a distance or at low resolution.Therefore, gait is very suitable for intelligent video surveillance. Actually,gait is very suitable for intelligent video surveillance. Actually, gaitrecognition is a hot research area now.At first, gait sequences are preprocessed, and the noise are removed.Then features are extracted from the image, by analyzing and comparingkinds of motion detection methods, and considering the background ofgait sequences, a moving object detection algorithm based on backgroundand the weights multiple interframe difference is presented. At first,background model is obtained by the method based on frame difference.Then, the moving object is extracted with background subtraction andmulti-frame-differencing, which is insensitivity to the target object’sspeed and environment disturbance. Finally, morphological gradientoperation is applied to remove the influence of outside noise. Final gaitcycle is analyzed, width and height of body analysis is performed tocomputer it.Based on the idea that joint-angle trajectories of body parts inwalking motion include sufficient dynamic identity information, a gaitrecognition method based on lower-limb motion analysis and dynamictime wrapping is proposed. For each gait sequence, according to theknowledge of human body anatomy, the coordinates of lower-limb jointsare obtained by analyzing lower-limb motion, and then the trajectories oflower-limb angles in one cycle are extracted as feature vectors. Dynamictime wrapping is used to measure the similarity of different sequences, then, nearest neighbor algorithm and K-neighbor algorithm are finallyperformed to realize gait recognition.However, gait is a behavioral characteristic. In comparison withother biometric features, it has inherent disadvantages, such as lowaccuracy, weak robustness. In order to improve the recognitionperformance of video surveillance system, multiple biometric featureinformation is used and a hybrid fusion recognition method is proposedusing face and gait. In the fusion process, Rank-Level ideology is adopted,and each feature is assigned to weights, which can make them combine insuitable proportion. Experimental results demonstrate that recognitionrate of proposed method is100%in a small scale database, much higherthan those based on single feature.