基于改进SSD神经网络和点云配准算法的变电站三维模型快速建立方法

Fast modeling method of 3D substation model based on improved SsD neural networks and point cloud registration algorithm

为提高变电站三维建模的效率,提出了一种改进SSD(single shot detection)目标检测算法和ICP(iterative closest point)点云配准算法结合的变电站三维模型快速建立方法。该方法针对变电设备点云训练样本数量不足造成的设备识别准确率低这一问题,将三维模型快速建模转换为设备类型、型号识别和点云导入。首先利用改进SSD目标检测算法对变电设备类型进行初步识别,然后运用ICP配准算法对变电设备进行型号识别,2种方法的结合实现了变电设备点云的准确识别,最后根据配准得到的设备在变电站点云场景中的实际位姿,将模型库中配准的标准模型导入变电站三维点云场景,极大地提高了变电站三维模型的建模效率。该方法已在某变电站三维建模中得到应用,结果表明,不同变电设备型号识别准确率较高,设备建模平均时间为32 s,效率远远高于人工建模。

In order to improve the efficiency of substation 3D modeling,a rapid modeling method of substation 3D model combining improved SSD(single shot detection)target detection algorithm and ICP(iterative closest point)point cloud registration algorithm is proposed.Aiming at the low accuracy of equipment recognition caused by insufficient number of point cloud training samples,the method quickly transforms 3D model modeling into equipment type,model recognition and point cloud import.First,the improved SSD target detection algorithm is used to initially identify the type of substation equipment,and then the ICP registration algorithm is used to identify the model of substation equipment.The combination of the two methods realizes the accurate identification of the point cloud of substation equipment.Finally,according to the actual pose of the registered equipment in the substation point cloud scene,the registered standard model in the model library is imported into the substation 3D point cloud scene,which greatly improves the modeling efficiency of the substation 3D model.The method has been applied to a substation 3D modeling.The results show that the model identification accuracy of different substation equipment is high,the average modeling time of equipment is 32 s,and the modeling efficiency is much higher than that of manual modeling.