3D Point Cloud Of Substation Equipment Recognition Based On Surface Features

With the development of GIS(Geographic Information System) towards the three-dimensional field, the three-dimensional digital substation attracts an increasing number of researchers’ attention and gradually becomes a hot topic in this filed. It is difficult to use three-dimensional laser scanner to obtain cloud data and recognize targets by three-dimensional components. Since the data collection environment is complex and variable, there must be noise, deletion, blocking, etc. in the actual scene. For this reason, the difficulty lies in the fact that there must exist difference among point cloud data, even though using the same type of equipment. As analyzed above, they are all the three-dimensional object recognition difficulties. To address these issues, this thesis mainly study the three-dimensional object recognition based on the three-dimensional point cloud data.The main work is as follows:(1) The spatial structure and hierarchy of the point cloud equipment is obtained through the octree method, so as to search neighboring points rapidly and improve the efficiency of data preprocessing.(2) During the streamline preprocessing, to conserve the original equipment’s surface information, adopting improved linear octree method makes streamlined data is from the original data on the device rather than the root node of the octree data.(3) The surface features, including the main curvature, Gaussian curvature, mean curvature as well as shape of the index values, can be extracted through the local parametric surface fitting method. By doing so, we can obtain the local features of the target surface.(4) We also study 3D object recognition method based on the HF voting. The target’s centroid position is acquired through establishing local reference frame, searching character and corresponding points, coordinating transformation and HF voting and target is recognized.(5) During recognition, a histogram of DSM is introduced in this thesis. It can save time and increase the efficiency of three-dimensional object recognition. Finally, several simulation experiments are conducted based on the three-dimensional object recognition algorithm, and the results show that the algorithm can effectively identify the three-dimensional point cloud device.(6) Some research are carried out to estimate the 3D posture based on principal component analysis method. Finally, the simulation results show that this method can accurately estimate the posture of three-dimensional equipment.