Amplitude spectrum-based gait recognition

Abstract

Gait is a biometric feature and identification of people from gait captured on video that has become a challenging problem in computer vision. A recognition method based on amplitude spectrum (Fourier spectrum) of frequency domain is proposed. Fourier spectrum reflects the frequency feature of person's pose in the current frame. After getting the period and key poses through the whole sequence according to the height of centroid and the amplitude value in frequency domain, we extract 5D features from the sequence including four amplitude values corresponding to four key poses image and the proportion of body. Finally, three measures are applied to evaluate our algorithm. Experiments result on USF dataset show that our method is insensitive to many changes condition of background, lighting, clothing, walking speed, segment noise and so on. For we implement gait recognition according to the discriminant analysis, the result to some extent is robust to the variation of viewpoint and cloth and recognition rate is satisfying.