Gait Recognition Based On The Spectrum Of The Procrustes Mean Shape

ABSTRACT:With the development of the society, the security issues become more and more concerned, especially the sites particularly sensitive to the security. So a lot of biometric technologies are being applied to the security. With its inherent advantages, such as non-invasive,long-distance identification,hidden-difficulty,more-acceptance, the gait has obtained more and more attention. Gait recognition is carried out according to the human gait to verification and recognition. With the continuous requirements of the security, the research of the gait will be more and more depth. The gait is more likely to be applied to all aspects of daily production and life. Most of the existing algorithms are based on the side view to recognition, the research purposes of the paper is to find a high recognition rate, high speed and multi-view gait recognition algorithm and build a recognition system based on the CASIA Database B to achieve and show the process of gait recognition. The work of the paper includes:Firstly, during the moving target detection, it applied a improved background subtraction algorithm, weighting the current frame and the background frame to get the new background image.Secondly, a new contour sampling method, dividing the body into three areas:head and shoulder area, torso area, legs and feet area. According the importance of the areas during the gait motion, the number of sampling points can be set flexibly in different regions.Thirdly, it is proposed a multi-view gait recognition algorithm based on the Fourier Spectrum Analysis of Procrustes Mean Shape (FSAOPMS) in this paper. The method of Procrustes shape analysis is used to produce Procrustes compact Mean Shape (PMS) from the continuous gesture variation of human body contours in gait sequences. Procrustes shape is the shape descriptor in the statistics, it is essentially a two-dimensional configuration vector. Analyze the spectrum of the PMS using the Fourier Transform. Make the amplitude spectrum as the finial feature.Fourthly, the amplitude spectrum was a real and scalar feature which was different from the vector feature and contained overall and detail contour information. It proposed to use the Euclidean distance to compute the similarity of the amplitude spectrum and the classifier are Nearest Neighbor (NN) and K Nearest Neighbor (KNN). The experiments on CASIA database showed that the proposed algorithm had got a satisfactory and encouraging result.Finally, on system construction aspect, in the paper it also constructed a simple gait recognition system based on the CASIA Database B. The system provided a platform for further gait recognition research. The system provided two indicators to determine the superiority of the algorithm, such as the correct recognition rate and the speed. The time consuming is counted according to the CPU frequency. The proposed algorithm was realized and verified in the system. The comparison with three other algorithms also realized in the system showed that the proposed algorithm had great advantage in both correct recognition rate and time consumption.