In recent years, a large number of video surveillance systems are deployed in the nature reserves, so it has become an urgent problem how to effectively use the increasing mass of video surveillance data. In this study, an efficient algorithm for key frame extraction based on image similarity is used to clean and compress the massive video data. At the same time, an object detection algorithm based on deep learning is used to extract valid video information. In addition, the system provides a variety of content-based video retrieval methods. It automatically analyzes and processes the search contents submitted by the user so as to quickly retrieve the video of interest. This study analyzes and retrieves the video surveillance data of wild animals in Qinghai Lake, which verifies the correctness of the proposed system.