基于2D&3D融合算法的扣件故障检测系统

Fastening Fault Detection System Based on 2D&3D Fusion Algorithm

扣件作为钢轨重要的紧固件之一,在列车运行过程中,会受到钢轨和枕木带来的巨大压力和振动。为保障扣件状态检测的准确性,研制了一套基于2D图像和3D深度信息融合的扣件故障检测系统。首先基于YOLOv4定位2D图中的扣件区域,提取出扣件检测区域;然后将深度图中对应的检测区域转换为点云图,采用点云匹配的方式对扣件状态进行判断;最后通过精度验证试验验证系统的检测精度。试验结果表明:系统整体的识别率为99.8%,误报率为1.68%;对于故障识别率,只有在扣件发生小偏转角度(≤5°)故障时,系统的识别率较低,其余设置的故障识别率均达到96%以上,能够满足工程化的要求。

As one of the important fasteners of rails,fastenings will be subjected to great pressure and vibration caused by rails and sleepers during train operation.In order to ensure the accuracy of fastening condition detection,a fastening fault detection system based on 2D image and 3D depth information fusion is developed.Firstly,the fastening area in the 2D map is located based on YOLOv4,and the fastening detection area is extracted.Then,the corresponding detection area in the depth map is converted into a point cloud map,and the state of the fastening is judged by point cloud matching.Finally,the detection accuracy of the system is verified by the accuracy verification test.The experimental results show that the overall recognition rate of the system is 99.8%,and the false alarm rate is 1.68%.For the fault recognition rate,only when the fastening has a small deflection angle fault(≤5°),the recognition rate of the system is low,and the fault recognition rate of the remaining settings is more than 96%,which can meet the engineering requirements.