Video-based person re-identification aims to retrieve specific pedestrians from videos taken by different cameras. However， it faces the problem of huge data volume and time redundancy of video data， that is， video data consume a lot of storage space and have strong correlation between different frames. Using all frames for identification will reduce the query efficiency， and the interference and noise in the video will also adversely affect the accuracy. In order to solve such problems， this paper proposes a key frame extraction algorithm for video-based person re-identification based on walking cycle clustering. Firstly， the algorithm extracts the candidate walking cycle by using the periodicity of the distance between feet of pedestrians. Then， it selects the key walking cycle as the key frame from the candidate walking cycle by clustering. Interference and noise are removed and only key frame information is used for identification to reduce the impact of time redundancy and improve accuracy. Finally， the algorithm is applied to video-based person re-identification， and the data will be processed before querying to reduce the storage space and to improve the query efficiency. Experimental results on MARS and DukeMTMC-VideoReID datasets show that the algorithm can reduce storages space by 59%—82%， the cumulative match characteristic Rank-1 is improved by 1.1%—1.4% and the mean average precision is improved by 0.2%—5%.