Review on key technologies of AI recognition for videos in coal mine
CHENG Deqiang, QIAN Jiansheng, GUO Xingge, KOU Qiqi, XU Feixiang, GU Jun, GAO Yachao, ZHAO Jinsheng
煤矿安全生产视频分析与识别技术是保障我国煤矿智能化建设和煤炭工业高质量发展的核心技术支撑。为及时对煤矿井下安全隐患进行实时监测和预警,视频AI(Artificial Intelligence,人工智能)识别关键技术已经成为煤矿安全生产领域的研究热点。阐述了我国煤矿智能化建设过程中安全监测与监控的发展现状,分析了当前矿井视频监控与安全隐患识别预警存在的效率低、响应慢、效果差等问题,结合计算机视觉、边缘计算、大数据处理、云服务、智能终端等先进技术手段、进行了煤矿安全生产视频AI 识别的顶层设计,提出了煤矿“人−机−环”全域视频AI 感知的“云−边−端”协同计算系统架构,构建了视频识别端节点传感器、边缘计算设备、视频识别场景云服务应用体系,明确了智能识别与预警联动控制响应机制,打通了“云−边−端”信息交互感知与联动控制数据链,实现了数据共享联动和预警协同。同时,围绕矿山“人−机−环”全域AI 视觉信息智能感知和全息泛化景象平台的构建,梳理了矿井安全隐患视觉感知及识别预警的技术处理流程,归纳了AI 识别过程中的各类预处理−增强−重建−检测−识别方法的优点和缺点,明确了煤矿安全生产视频AI 识别关键技术发展的主流方向和趋势。其次,结合王家岭煤矿、鲍店煤矿等代表性矿井的应用案例,示范展示了煤矿安全生产过程中实际典型应用场景等方面的最新进展和应用效果。最后,针对煤矿安全生产视频AI 识别关键技术的特点,总结了现有煤矿安全生产视频AI 识别系统存在技术理论薄弱、智能终端规格不一且应用场景混乱、数据兼容性及联动闭环能力较差、数据库安全性较弱、评价机制不统一、应用标准不完善等问题,指明了未来的发展方向是加强对视频AI 识别关键技术及理论的研究,建立健全智能终端硬件规格及适用体系,构建标准统一、机制完善、实时互联、动态预测、协同控制、安全可靠的煤矿信息多维度主动感知新模式和工业互联网应用平台,逐步形成全矿井全息泛化的高精度智能感知场,实现对井下“人−机−环”全域视频信息的精准感知和危险源协同管控。
The video analysis and identification technology of coal mine safety production is the core technical support to ensure the intelligent construction of our country's coal mines and the high-quality development of the coal industry. In order to carry out real-time monitoring and early warning for potential safety hazards in coal mines, the key technologies of video AI (Artificial Intelligence) identification have become the research hotspot in the field of safety production in coal mines. In this paper, the development status of safety monitoring in the process of intelligent construction of coal mines are first expounded. Then, the problems of low efficiency, slow response and poor effect of the current mine video monitoring and safety hazard identification as well as early warning system are concluded. Combined with advanced technologies such as computer vision, edge computing, big data processing, cloud services, and intelligent terminals, the toplevel design of AI recognition for coal mine safety production video is carried out. Furthermore, the “cloud-edge-terminal” collaborative computing system architecture of “human-machine-environment” global video AI perception in coal mines is also proposed, followed with a video recognition end node sensor, edge computing equipment, and video recognition scene cloud service application system constructed. By this way, the intelligent identification and early warning linkage control response mechanism are clarified, and the “cloud-edge-terminal” information interactive perception and linkage control data chain has been dredged, resulting with data sharing linkage and early warning coordination. At the same time, around the construction of the “human-machine-environment” global AI visual information intelligent perception and holographic generalized scene platform, the technical processing process of visual perception and identification and early warning of mine safety hazards has been sorted out. What’s more, the characteristic of the processing-enhancement-reconstruction-detection-recognition method are also summarized, and the mainstream direction and trend of the key technology development of coal mine safety production video AI recognition are also pointed out. Secondly, based on the application cases of representative mines such as Wangjialing Coal Mine and Baodian Coal Mine, the author demonstrates the latest progress and application effects of typical application scenarios in the process of coal mine safety production. Finally, according to the key technology characteristics of coal mine safety production video AI recognition, it is concluded that the existing coal mine safety production video AI recognition system has weak technical theory, different specifications of intelligent terminals, confusing application scenarios, poor data compatibility and linkage closed-loop ability, weak database security, inconsistent evaluation mechanism as well as imperfect application standards, etc. Subsequently, this paper pointed out that the future development direction is to strengthen the research on key technologies and theories of video AI recognition, establish and improve intelligent terminal hardware specifications and applicable systems and build a new coal mine information multi-dimensional active perception model and industrial internet application platform with unified standards, perfect mechanism, real-time interconnection, dynamic prediction, collaborative control, safety and reliability, which gradually form a high-precision intelligent perception field of holographic generalization in the whole mine, so as to realize the precise perception of the underground "human-machine-environment" global video information and the coordinated control of danger sources.
safety production in coal mines; video AI recognition; system architecture; image enhancement; object detection
主办单位:煤炭科学研究总院有限公司 中国煤炭学会学术期刊工作委员会