Study On Gait Recognition Based On Moment And BP Neural Network

With the development of the security requirement of information society, it becomes the mainstream to identify one's identification with biometric character today. The biometrics recognition has unexampled advantages compared with previous conventional figure identifying method. Especially for gait recognition, it has more extensive predominance such as unaffecting, simple collection, non body-invading, far-away-identification, and difficult to be disguised. So it becomes a hot spot and more attractive by many computer vision researchers. This paper has studied gait recognition technology on four points: gait detection, gait representation, feature analysis and gait classification. And, the gait representation and feature analysis are the most important.(1) In the paper, background subtraction is used in gait detection. We use morphological processing to clean off holes and noise. And then the image is normalized. Finally, we get the gait contour precisely by contour tracking.(2) The gait representation based on moment is discussed. A novel gait representation method based on the feature of image moment is presented in the paper. We extract Hu moments and Zernike moments for gait representation. To improve the practicality and reduce the complexity of the algorithm, after comparing the difference of Hu and Zernike moments during experiments, we get six Zernike moments by the method of scatter criterion with sequential forward selection and make them to be the features we adapt in the following steps.(3) A gait recognition algorithm based on BP neural network is presented in the paper. Because of the particularity of gait recognition, we design a BP neural network which can work effectively based on theory analysis and experiments. The experiment result shows that the algorithm is effective and we have got higher recognition rate.