基于深度学习的电力作业人员行为识别技术

Behavior Recognition Technology of Power Operating Person Based on Deep Learning

摘要:

目前的电力现场安全监控主要通过人员对监控视频进行全程监测，但采用人工检测的方法不仅浪费时间，而且容易出现漏报的情况，使工作人员的人身安全无法得到保障。为实现对作业现场人员行为的智能识别，提出了一种基于OpenPose的电力作业人员的危险行为识别技术。该方法提取视频流图像中电力作业人员骨骼关键点信息，利用深度神经网络实现多人场景下电力作业人员的人体行为姿态感知，实时对施工人员的违规行为进行检测识别，并发出警告。所提方法实现了对电力现场作业人员行为的准确、实时安全监控，保障了现场作业人员的人身安全和电力作业的顺利进行，具有一定的鲁棒性与泛化能力。

Abstract:

At present, the field safety monitoring relies on the professionals to monitor the surveillance video throughout the process, and the manual detection method is very timeconsuming and prone to false alarms, which is difficult to ensure the personal security during the operation. In order to realize the intelligent recognition of human behaviors in power operation field, an OpenPosebased method for detecting dangerous behaviors of electric power workers is proposed. This method extracts the key point information of electric workers′ skeletons from the video stream images, and uses deep neural network to realize human behavior situation awareness in multiperson scenarios, which can realize the realtime detection of construction personnels′ illegal behavior and issue early warning. The proposed method realizes the accurate and realtime safety monitoring of human behaviors in power operation field, and guarantees the personal security in the field and the smooth progress of power operation, and the model has a good robustness and generalization ability.