自然资源视频监测目标检测关键技术研究与应用

Research and Application of Target Detection Key Technologies for Video Monitoring on Natural Resources

随着机器视觉技术的突飞猛进,智能视频监测为自然资源监管提供了可实时响应的新执法手段。运用基于深度学习的域自适应目标识别技术和基于DEM辅助的高精度单目定位技术,对自然资源违法行为与线索进行快速目标检测;并在此基础上设计了自然资源视频执法监测系统技术架构,在广东开展了实地应用测试。结果表明,该方法对重型机械、土堆、建筑板材、在建砖房、新建板房、红砖堆等类别的识别准确率均高于80%,在1 200 m监测距离内平均定位误差小于3%。

With the rapid development of machine vision technology,intelligent video monitoring provides a new law enforcement means that can respond in real time for natural resource supervision.In this paper,we used domain adaptive target recognition technology based on depth learning and high-precision monocular positioning technology based on DEM assistance to rapid detect the illegal acts and clues of natural resources.On the basis of this,we designed the technical framework of natural resources video law enforcement monitoring system,and carried out field application tests in Guangdong.The results show that the accuracy of this method in identifying heavy machinery,mounds,building boards,brick houses under construction,new plank houses,red brick piles and other categories is higher than 80%,and the average positioning error is less than 3%within the monitoring distance of 1200 m.