摘要
由于超载、过度使用和缺乏维护等原因,道路表面容易损坏出现坑槽。提出了一种使用廉价的深度摄像头的基于深度学习的沥青路面坑槽量化方法。首先,根据检测到的道路坑槽的位置和深度信息,使用RANSAC算法识别并分割道路表面。然后,创建了带标签数据库,利用Faster R-CNN进行训练、验证和测试。该方法可以不受传感器与检测目标之间距离的影响,实现路面平面的拟合,并依靠RGB-D传感器的输出来识别和量化多个道路坑槽体积。最终结果表明,模型AP值可达90.79%,且对于单个坑槽测量体积,平均精度误差值低于10%。
Due to overload,overuse and lack of maintenance,the road surface is prone to potholes.This study proposes a deep learning-based method to quantify potholes in asphalt pavements using cheap depth cameras.First,according to the location and depth information of the detected road potholes,the RANSAC algorithm is used to identify and segment the road surface.Then,a labeled database was created,and Faster R-CNN was used for training,verification and testing.The method can fit the road surface without being affected by the distance between the sensor and the detection target.In addition,it relies on the output of the RGB-D sensor to identify and quantify the volume of multiple road potholes.The final result shows that AP value of the model can reach 90.79%,and for individual potholes,the mean accuracy error value is less than 10%.
作者
余俊
吴海军
王武斌
张宗堂
YU Jun;WU Haijun;WAGN Wubin;ZHANG Zongtang(Modern Rail Transit Application Technology Research Center,Chengdu Vocational and Technical College of Industry,Chengdu,Sichuan 610000,China;Hunan Commnications Research Institute Co.,Ltd.,Changsha,Hunan 410015,China;National Engineering Laboratory for Technology of Geological Disaster Prevention in Land Transportation,Southwest Jiaotong University,Chengdu,Sichuan 610000,China;Geotechnical Engineering Stability Control and Health Monitoring Hunan Provincial Key Laboratory,Hunan University of Science and Technology,Xiangtan,Hunan 411201,China)
出处
《公路工程》
2022年第4期95-102,共8页
Highway Engineering
基金
国家自然科学基金项目(51909087)
2018年度中铁建科技重大专项(2018-A01)
湖南科技大学岩土工程稳定控制与健康监测省重点实验室开放基金资助项目(E21807)。
