Human abnormal behavior detection method based on T-TINY-YOLO

Abstract

Aiming at the problem of human abnormal behavior detection in video surveillance, an abnormal target detection method based on T-TINY-YOLO network model is proposed. First, the type of abnormal behavior is defined according to the requirements of the monitoring scenario. Then the calibrated abnormal behavior data is trained through the YOLO network model to achieve end-to-end abnormal behavior classification, thereby achieving abnormal target detection for specific application scenarios. For the characteristics of a large number of zero-valued weight parameters in YOLO's network weights, a convolutional neural network tailoring scheme is proposed to improve the network model, accelerate the algorithm, and improve the real-time performance of the system. Finally, the method is implemented on the embedded platform NVIDIA JetsonTX2. The experimental results are good, the detection speed reaches 12 frames/s, and the recall rate is 80.87%.