基于轻量级YOLOv3的员工违规行为检测算法

A Staff Violation Detection Algorithm Based on Lightweight YOLOv3

随着工业互联网安全受到越来越多的关注,如何通过技术手段保障生产环境安全就成为单位管理人员所思考的问题。论文针对值班室、中控室、管理室等类似场景的安全需求,提出了一种基于轻量级卷积网络的员工违规行为检测算法。该算法能够实时监测指定区域工作人员的行为状态,避免因员工长时间处于离岗、睡岗状态时,可能导致设备设施生产事故、设备信息泄露、无法处理应急事件等情况的发生。违规行为检测算法由人体检测与行为识别算法两部分组成。首先,通过轻量级人体检测网络获取人体检测框;而后,利用目标跟踪算法与行为识别算法对人体检测框进行识别,进而确定工作人员是否存在违规行为。实验数据表明:该算法大大减少网络权重以及计算量,在边缘设备Hi3559A上检测速度可达13 ms,在实际场景数据集上,违规行为检测准确率可达96.6%。

With industrial Internet security receiving more and more attention,how to ensure information security management through technical means has become a question for unit managers to consider.In this paper,a lightweight convolutional network-based algorithm for staff violation detection is proposed for the security needs of similar scenarios such as duty rooms,central control rooms and management rooms.The algorithm is able to monitor the behavioural status of staff in designated areas in real time,avoiding malicious incidents such as production accidents in equipment and facilities and leakage of equipment information that may occur when staff are away from work or sleeping for long periods of time.The violation detection algorithm consists of two parts,which are the human detection and the behaviour recognition algorithm.Firstly,the human detection frame is obtained through a lightweight human detection network.The human detection frame is then identified using a target tracking algorithm and a behaviour recognition algorithm to determine if there is a breach of the rules by the staff member.Experimental data shows that the algorithm significantly reduces network weights as well as computational effort,with detection speeds of up to 13 ms on edge devices and violation detection accuracies of up to 96.6%on real-world scenario datasets.