【作者】 梁伟； 荆朝； 周治国； 姚楠； 王益男；

【Author】 LIANG Wei;JING Zhao;ZHOU Zhiguo;YAO Nan;WANG Yinan;Research Institute,State Grid Jiangsu Electric Power Co.,Ltd.;School of Information and Electronics,Beijing Institution of Technology;

【机构】 国网江苏省电力有限公司电力科学研究院； 北京理工大学信息与电子学院；

【摘要】 电力施工场景中,安全帽对施工者的人身安全有着重要保护作用。因此,通过监控摄像头实现对安全帽状态的智能检测对施工安全至关重要。首先采用轻量级深度学习神经网络,从图像中充分提取不同尺度特征,并生成特征图;再通过卷积预测器,进行对多尺度特征分析,完成图像中对人和安全帽目标的检测。实验证明,该算法能够在复杂施工场景中精确的检测出人和安全帽佩戴状态,且在边缘计算板卡上也体现出了较好的实时性。更多还原

【Abstract】 In electric construction scene,wearing a helmet plays an important role in protecting the worker.Therefore,the intelligent detection algorithm for helmet and worker by cameras is crucial.This paper uses a small convolutional neural network to fully extract different kinds of features from the image.Then it uses the convolution detector to analyze the features to predict the classification and location of people and helmet targets in the images.Experiments show that the algorithm can accurately detect multi-scale targets in complex construction scene,and this algorithm also presents good real-time performance on edge computing platform.更多还原