Research On Algorithm Of Human Gait Recognition Based On Sparse Representation

With the rapid development of pattern recognition, computer vision and digital signal processing technology, biometric technology has received extensive attention by researchers. From the early fingerprints, palm prints and DNA identification to recently face recognition and iris recognition, they have been widely applied in various of fields, such as security, financial, identification and so on. Although recognition technologies which have been mentioned are gradually mature, the manners described above require not only the active cooperation of people, but also the close contact, even direct contact. Therefore, gait recognition has become a hot research interests recently.Compared with the above biometric technologies, gait recognition has the irreplaceable advantages, such as non-contact, non-invasive, difficulty to disguise and hide. The researchers in computer vision and pattern recognition fields have been greatly interested in it. However, due to environment, perspective, shelter materials and shooting conditions, they are challenging to gait recognition. Gait recognition technique cannot still meet the strict requirements for real-time and accuracy of practical applications. In this paper, we present a novel sparse representation based gait recognition algorithm. We perform the experimental validation on CASIA gait database and USF database to evaluate the performance of the proposed method and compare it with the state-of-the-art gait recognition methods. The main research works are listed below:First, we perform image preprocess for the human walking video. For the image of each frame, we use the human target detection method to extract silhouette image of the human, and finally obtain the silhouette image sequence of the walking human. Then, through the study of foot posture, we calculate the gait period of the silhouette image sequence, preparing for subsequent feature extraction.Moreover, we extract Gait Energy Image(GEI) by averaging gait silhouettes across a gait cycle. Then we adjust the horizontal centers of different areas of walking human based on GEI and shift the corresponding areas to obtain Shifted Energy Image(SEI). Because random noise was suppressed in the process of gait image averaging, so the robustness of GEI is greatly enhanced, and GEI contains static and dynamic characteristic information of gait. For the SEI image, the effective shift of head, torso and legs helps reduce the influence on the feature representation caused by local movement. The next, Gabor filter is used to extract the multi-scale and multidirectional spatial frequency features of SEI. In this study, we use the 8 direction and 5 scale Gabor filter functions to extract the amplitude spectrum of the Gabor filter response on the SEI image and enhance edge characteristics. Then, we use local binary patterns on the Gabor image to get local Gabor binary patterns and perform histogram statistics.Finally, we utilize the sparse representation technique to perform gait recognition. The experimental results on USF Human ID database and CASIA gait database clearly demonstrate the efficacy of the proposed method.