A Bag-of-Gait Model for Gait Recognition

In this paper, we propose a novel gait recognition method based on a bag-of-gait model. In the proposed method, the image sequence of a walking person is encoded by a codebook consisting of a list of code words denoting different walking stages; then, this image sequence is represented by a feature vector denoting the existence of the code words, which is further used for classification. Unlike most of previous gait recognition methods, there is no need to estimate gait period in the proposed method. Moreover, the proposed method is capable of recognizing the gait when the observed gait period is incomplete which caused by occlusion or short appearance time. The method is evaluated on a dataset consisting of 151 subjects with four different walking conditions using four-fold cross-validation. The result shows the effectiveness of the proposed method and the state-of-art result is achieved on that dataset.