Memory-based Gait Recognition

In this paper, inspired by the mechanism of memory and prediction in our brains [2], we propose a straightforward and effective memorybased gait recognition method (MGR) to realize the memory and recognition process of the gait sequences. Because of various covariates including carrying, clothing, surface and view angle, we extract the robust 2D joint location information via the joint extraction model as the gait features. Compared to the traditional neural network, the memory neuron network (MNN), for example, the Long Short-term Memory (LSTM) architecture, simulates the human brain and stores the objects in the weights of neural connections. Besides, by the large-scale parallel computing, MNN can repair the incomplete and tainted data (the extracted 2D gait feature is dirty). It is the first time that we utilize the MNN to address the gait recognition issue. This maybe empower a fresh orientation for solving gait recognition problem. Fig.1 shows the overall framework of the method.