关键词: 目标检测, 行为分析, 深度学习, 目标数据集, 图像处理

Abstract:

Aiming at the problem of human behavior analysisin key positions, a video-based behavior analysis method is proposed. A personnel behavior dataset containing multiple pose samples is produced, which is trained by the YOLOv3 network to obtain a behavior detection model. The proposed human behavior analysis algorithm combined with the model is used to process video and conduct a preliminary analysis of human behavior. Based on the human behavior analysis algorithm, combined with image similarity, brightness and darkness and other characteristics, in-depth analysis is carried out and the judgment results of leaving, sleeping and playing mobile phone are given. The experimental results show that the produced dataset has higher detection accuracy in human behavior detection, and the accuracy of behavior analysis is extremely high, and it can be processed in real time.

Key words: object detection, behavior analysis, deep learning, object dataset, image process