Examination room abnormal behavior analysis method and system based on deep learning, and terminal

The invention provides a deep learning-based examination room abnormal behavior analysis method and system and a terminal, and relates to the field of video behavior analysis, and the method comprises the steps: obtaining examination room monitoring video information, and carrying out the preprocessing of the examination room monitoring video information; constructing a convolutional neural network, and performing position detection on invigilators and examinees in the monitoring image; tracking invigilators and examinees in the monitoring image in real time; by using the constructed convolutional neural network, taking one branch behind the decoupling head to carry out behavior classification identification, and identifying two behaviors of sitting and standing; judging whether the individual is a student or not according to the identified behavior category and the standing duration and displacement condition of the individual under the continuous frames; and if the student stands, the system is triggered to