基于时空双流的矿井人员求救动作识别  

Identification of Rescue Actions by Mine Personnel Based on Spatiotemporal Dual Flow

作　　者：杨世超 YANG Shichao(Anbiao National Center for Mining Products Safety Sign,Beijing 100000,China)

机构地区：[1]安标国家矿用产品安全标志中心有限公司,北京100000

出　　处：《煤炭技术》2025年第1期175-177,共3页Coal Technology

摘　　要：当矿井人员遇险通过动作发出求救信号时,救援系统若能及时准确识别矿井人员求救动作,则能快速响应,实施救援,从而避免人员伤亡事故发生或降低事故严重程度。研究了煤矿井下视频中人员求救动作识别问题,提出了一种特征通道注意力及多空间注意力的时空双流动作识别解决方案,该方案的特点是融合了特征通道信息的差异性与空间中多个运动显著区域的特征,实验验证该方案的求救动作识别准确率达92.8%。When mine personnel encounter danger and send out distress signals through actions,if the rescue system can timely and accurately recognize the rescue actions of mine personnel,it can quickly respond and implement rescue,thereby avoiding the occurrence of casualties or reducing the severity of accidents.The research studied the problem of identifying personnel's distress actions in underground coal mine videos and proposed a spatiotemporal dual stream action recognition solution based on feature channel attention and multi spatial attention.The feature of this solution is the fusion of the differences in feature channel information and the characteristics of multiple significantly moving areas in space.The experimental verification shows that the accuracy of distress action recognition in this scheme reaches 92.8%.

关 键 词：矿井人员 动作识别 时空双流 注意力机制 

分 类 号：TD76[矿业工程—矿井通风与安全]