基于深度学习的电网监控视频中工作人员检测与识别  

Detection and identification of staff in power grid monitoring video based on deep learning

作　　者：刘颖 胡楠 杨壮观 同东辉 胡畔 LIU Ying;HU Nan;YANG Zhuang-guan;TONG Dong-hui;HU Pan(Information and Communication Branch,State Grid Liaoning Electric Power Co.Ltd.,Shenyang 110006,China)

机构地区：[1]国网辽宁省电力有限公司信息通信分公司

出　　处：《沈阳工业大学学报》2019年第5期544-548,共5页Journal of Shenyang University of Technology

基　　金：国家自然科学基金资助项目(51307051)

摘　　要：针对电网监控视频场景多样,电网工作人员姿态变化严重影响工作人员识别精度的问题,提出了一种基于深度学习的电网监控视频中工作人员检测与识别算法.该算法使用ResNet50网络提取行人特征,Faster-Rcnn检测方法快速、精确地检测出电网中的工作人员,识别网络对检测出的工作人员进行身份确认,并使用各种组合损失来训练检测与识别网络.在电网监控视频数据集上的测试结果表明,所提出的方法具有更高的检测和识别精度,且对遮挡及低光照图片具有较好的鲁棒性.Aiming at the problems that the scenes of power grid monitoring video are diverse and the posture change of power grid staff seriously affects the accuracy of staff identification,a detection and recognition algorithm for the staff in power grid monitoring video based on the deep learning was proposed.The pedestrian features were extracted with the ResNet50 network in the proposed algorithm,and the staff in power grid were quickly and accurately detected with the Faster-Rcnn detection method.In addition,the identification network was used to identify the detected staff,and various combined-losses were used to train the detection and identification network.The test results of data set composed of power grid monitoring videos show that the as-proposed method has higher detection and recognition accuracy and better robustness to occlusion and low-light pictures.

关 键 词：监控视频 工作人员 行人识别 行人检测 深度学习 ResNet50网络 损失函数 Faster-Rcnn检测方法 

分 类 号：TM[电气工程]