Video-based deep learning parabolic behavior detection method

The invention relates to a video-based deep learning parabolic behavior detection method, and the method comprises the steps: making a data set needed by parabolic behavior detection for a deep learning selection framework, carrying out the algorithm selection and parameter adjustment of deep learning, and carrying out the iteration and training of a model in the framework, and obtaining a deep learning parabolic behavior detection model; and mounting the model to a video stream detection system, locking a parabolic person through the video stream detection system and a distance and angle analysis algorithm, calculating a distance and an angle, judging a parabolic level, and uploading to an enterprise data management platform. In the whole process, manpower and time cost are saved, the misjudgment rate is low, as long as certain original data exist, the needed model is automatically analyzed, learned, extracted and iterated, and then nationwide large-scale expansion is easily achieved. Meanwhile, the paraboli