The invention provides a shared bicycle quantity statistics and density estimation method based on computer vision, which comprises the following steps: acquiring a large number of pictures containingshared bicycles in different scenes, labeling and generating a data set; further processing the data set, and converting the original bicycle distribution map into a corresponding density distribution map by adopting a Gaussian function; dividing the data set into a training data set and a test data set; constructing a multi-column CNN model; and taking the training data set as the input of a CNNmodel, calculating a loss function by using the correspondingly output density estimation map and the previously marked density distribution map, and carrying out back propagation on the training model; deploying the trained CNN model into a video analysis system, analyzing video stream data collected by video monitoring equipment, and detecting whether the number and density of the shared bicycles exceed standards or not. According to the invention, a deep learning technology is used. On-site conditions and information can be completely presented, and the method is highly intelligent and canstably operate in a visual analysis system for a long time.