A Method for Analyzing and Warning the Safety Behavior of Smart Campus Personnel Based on Deep Learning

Abstract

To improve the intelligent application of products such as safety behavior warning, behavior sign collection, and abnormal point recognition management for campus staff, this article proposes an emerging video analysis technology method based on the development of computer vision technology. Firstly, the presence of human bodies in the area is determined through object detection algorithms, and then the key points of human bones are detected through pose estimation networks. We conduct in-depth analysis based on human key point action behavior analysis algorithms, compare them with human action rules that require warning behavior, and then determine whether the action has dangerous behavior, to achieve the warning effect. The experimental results show that the AP (Average Precision) value of the target detection algorithm reaches 0.967, and the AP value of the attitude estimation network model reaches 0.897. Therefore, combining these two methods can achieve action judgment and achieve early warning function, greatly reducing labor costs, and improving security awareness.