The invention discloses a human body abnormal behavior detection method based on a T-TINY-YOLO network, and belongs to the technical field of image and video analysis and processing in computer vision. The method comprises the following specific steps: firstly, according to different monitoring scenes, selecting a video sequence within a period of time, converting the video sequence into a picture, storing and preprocessing the picture; labeling four human abnormal behaviors which comprises finger, push, hug and standing in each picture by using a labeling tool, and generating an xml file as adata set; then, dividing the data set into a training sample and a verification sample, and inputting the training sample and the verification sample into an improved T-TINY-YOLO network model for training and verification; and finally, for a new monitoring video frame picture, directly inputting the new monitoring video frame picture into the trained T-TINY-YOLO network model after preprocessing, and outputting the category to which the calibration result of the human abnormal behavior belongs, thereby realizing end-to-end abnormal behavior classification. According to the invention, the size of the network is cut, the problem of network redundancy is solved, and the algorithm is accelerated, so that the network detection and time utilization are more effective.