摘要
针对以人为中心的井下视频监控模式存在持续时间受限、多场景同时监视困难、人工监视结果处理不及时等问题,提出了基于深度卷积神经网络的井下人员目标检测方法。首先将输入图片缩放为固定尺寸,通过深度卷积神经网络操作后形成特征图;然后,通过区域建议网络在特征图上形成建议区域,并将建议区域池化为统一大小,送入全连接层进行运算;最后,根据概率分数高低选择最好的建议区域,自动生成需要的目标检测框。测试结果表明,该方法可以成功检测出矿井工作人员的头部目标,准确率达到87.6%。
In view of problems that human-centered video monitoring mode had limited duration,multiple scenes were difficult to monitor at the same time,and results of manual monitoring were not processed in time,target detection method of underground personnel based on deep convolutional neural network was proposed.Firstly,input image was scaled to a fixed size,and a feature map was formed after operation of deep convolutional neural network;then,a suggestion area was formed on the feature map through area suggestion network,the suggestion area was pooled into a unified size which was sent to full connection layer for operation;finally,the best suggestion area was selected according to probability score,and the required target detection box was automatically generated.The test results show that the method can successfully detect head of underground personnel with an accuracy rate of 87.6%.
作者
唐士宇
朱艾春
张赛
曹青峰
崔冉
华钢
TANG Shiyu;ZHU Aichun;ZHANG Sai;CAO Qingfeng;CUI Ran;HUA Gang(School of Information and Control Engineering,China University of Mining and Technology, Xuzhou 221008,China;College of Computer Science and Technology,Nanjing University of Technology,Nanjing 211816,China)
出处
《工矿自动化》
北大核心
2018年第11期32-36,共5页
Journal Of Mine Automation
基金
国家自然科学基金项目(51574232)
