RGB pixel information fusion segmentation of dense point cloud data for transmission and distribution 3D digital corridors

Point cloud segmentation, as a key link in 3D point cloud data processing, can realize power transmission and distribution positioning, object identification, classification, and feature extraction, which helps to refine power grid management. In this paper, firstly, dense point cloud transmission and distribution 3D digital corridor modeling is carried out. Alignment splicing and noise reduction are carried out after obtaining the original dense point cloud. Contour line extraction, geometric modeling, and texture mapping are realized after processing the data to ultimately realize the transmission and distribution of 3D digitization. Then, the conversion formula for the pixel coordinate system and world coordinate system is derived to extract features from point clouds. Finally, a distance-based feature fusion method is designed to extract spatial features from point clouds and use the joint attention layer to segment them by fusing RGB pixel information. The original dense point cloud of a transmission and distribution digital corridor is segmented using the model presented in this paper for application after testing the dataset. It is found that the under-segmentation ratio of this paper’s algorithm is 0.96%, 3.44%, and 2.87% for the three scenarios of regular, irregular, and multi-targets, respectively, which is much lower than that of RANSAC+ECS with FCM + ECS. The intersection and concatenation ratios of this paper’s algorithm for the scenarios of irregular geometry as well as multi-target objects are 91.49% and 89.56%. It is much higher than 64.31% and 72.17% for RANSAC + ECS and 76.85% and 60.91% for FCM + ECS, which illustrates that this paper’s algorithm has a significant advantage in performance. In this study, the target point cloud can be segmented with high accuracy from the dense point cloud of a 3D model of power transmission and distribution with a large amount of data, effectively avoiding the phenomenon of under-segmentation and over-segmentation and contributing to the accurate control of power grid data.