基于深度学习和椭圆检测的城市道路积水深度监测方法研究

Research on Monitoring method of Urban Road Waterlogging Depth Based on Deep Learning and Ellipse Detection

为了解决传统城市内涝监测方式大量耗费人力物力,成本较高,不能满足城市内涝全面快速监测需要的问题,提出一种利用深度学习技术和椭圆检测算法的城市道路积水深度监测方法,通过深度学习模型对不同类型车辆的车轮进行检测和分割,利用椭圆检测算法对淹没车轮的几何特征参数进行提取,从而构建道路积水深度计算模型。通过东营市典型视频监控站点进行验证,结果表明:模型对数据集的平均定位精确率和分割精确率可达到94%以上,在实际积水监测中模型对正侧面和斜侧面车辆均具有较好的积水深度识别效果,近点处的识别结果优于远点,对正侧面车辆的识别结果优于斜侧面车辆。研究成果可为相关研究的进一步开展做铺垫,为城市内涝监测和洪涝灾害应急管理提供技术支撑。

In order to solve the problem that the traditional urban waterlogging monitoring methods consume a lot of manpower and material resources and also have a high cost,which cannot meet the needs of comprehensive and rapid monitoring of urban flood,a method of monitoring the depth of urban road waterlogging using deep learning and ellipse detection algorithm is applied,which constructs a computation model of the urban road waterlogging depth by detecting and segmenting the wheels of different types of vehicles on the images through a deep learning model and using ellipse detection algorithm to extract the geometric characteristic parameters of submerged wheels.Verified by typical video monitoring sites in Dongying City,The results show that the average positioning precision and segmentation precision of the model on the dataset can reach over 94%;the model has a good monitoring effect on waterlogging depth for vehicles on both the directly lateral side and the oblique lateral side in actual waterlogging monitoring;furthermore,the results at the near point are better than that at the far point,and the results for vehicles on the directly lateral side are better than the oblique lateral side.The results can lay the foundation for further research,and provide technical support for urban waterlogging monitoring and flood emergency management.