Research And Implementation Of The Key Technologies In Gait Recognition

Recognition by gait is a new field in biometric identification. Gait refers to the walk style. As one of the most attractive research area in computer vision, gait recognition is getting a wide range of applications. Gait recognition based on motion vision is used to identify the individuals in the video sequence, it integrates technologies as video sequence processing, pattern recognition and computer vision etc. The key technologies in gait recognition are pattern extracting and classification methods. This thesis probes into gait recognition with the videos as input in both theoretical and practical perspective. The main contributions of this thesis are listed as follows:1) Studied the preprocess methods of gait recognition, the following steps are preceded:First of all, establish the background model using median value method, get the silhouette by subtracting the sequence with the model. Get the region of human body according to the horizontal and vertical projection and have it normalized. Average the binary contours to obtain the GEI(Gait Energy Image) for further use in the feature extraction.2) Proposed the approach employing the Modular-(2D)2PCA to extract the feature for classification and using NN(Nearest Neighbor) method for recognition, and applied it to USF gait database. Experimental results demonstrate that the proposed algorithm performs an encouraging recognition rate and robustness against changing of clothes and carrying bags, and largely reduced the computation dimensions.3) Researched on the application of invariant moment in gait recognition, realized Zernike invariant moment and wavelet moment, which have features of translation, scale and rotation invariance. Then developed a hybrid moment to descript the gait characteristics. SVM(Support Vector Machine) is performed in recognition. Experiments on UCSD gait database achieved good results and proved its effectiveness.