Abstract: In order to solve the problem that convolutional neural networks（CNN） occupies too many memory resources in the implementation of lower computer in gait recognition, resulting in poor real-time performance, a gait recognition system based on lightweight convolutional neural network with STM32 single chip microcomputer as the hardware core was established. By building a gait recognition model based on lightweight convolution neural network, and using the methods of weight clustering and Huffman coding, the key parameters of convolution layer and full connection layer in the gait model based on lightweight convolution neural network were compressed, so as to reduce the amount of calculation and memory occupation of lower computer, so as to improve the real-time performance of the gait recognition system. The experimental results show that in terms of real-time performance, the gait recognition system based on lightweight convolutional neural network needs 375 ms to recognize the current gait, which reduces the data delay of 325 ms compared with the gait recognition system without model compression. In terms of accuracy, the recognition rate of gait recognition system in five gait modes of walking, going upstairs, going downstairs, going uphill and going downhill has reached 95.79%.