SpacePower is a visual toolkit to process airborne lidar pointcloud data specialized for UHV powerline satety evalution. The Power Grid Company would like to monitor their UHV power transmission line fast and low cost. They trend to use UAV to substitute manual patrolling. As UAV based lidar scanning is a very mature technique, the main challenge is how to process scanning data efficiently and accurately. This is design goal of SpacePower.
I developed several method/algorithm to fulfill the pipeline. These two video shows the workflow of SpacePower.
Pylon detetion or pylon auto-localization is the first step of analysis. Pylons must be recognized in unorganized point cloud. I designed algorithm based on 2D features on XOY coordinate framework. Point clouds are project to XOY plane, then some features would be computed based on points of each cell. Many channels of feature would be fused to describe powerline pass-through region. Morphology operations are taken to refine the mask extracted. Finally, we find local maximum point in density channel as the location of pylon.
Power lines are the most inspecting valuable objects in power corridor, how to extract it without missing and deformation is most important question. At here, we designed block wised histogram splitting method.
Catenary is the curve that an idealized hanging chain or cable assumes under its own weight when supported only at its ends. Fitting catenary model requires pretty well initial esitmation and purely point data. First we use linear region grow algorithm to cluster each bundles of line. Then give the initial estimation from particular points on the line. Finally, use LM solver find the optimized solution.
Hanging Point is the joint of line and insulator. Recognize insulator from LiDAR Point cloud is pretty difficult, while hanging point is much easier as we can infer from lines. WE designed section area tracking method to deal it.
As said before, interactive operation is neccessary. Pylon inclination calculation is the typical one. It’s very hard to analyze pylon pattern from raw point cloud. We tried other method like recognize horizontal sections, could not reach ideal performance. Alternatively, we take manipulator as one of procedures. Their work is to pick 4 brace one-by-one. Then algorithm would interpolate pylon’s centre axis.
Visualization is the last but not least part of the diagnosis. We decided to display diagnosis result from several ways, such as diagram, profile, and 3D scene layer. The part of working is pending, here is the preview of this module.
I’ve taken an important role in this project, as a Coordinator of several subsystems and leader of LiDAR diagnose system. Actually, the whole subsystem is writen by me. From architecture, design, experiment, and implement.
During the period of this project, including now. I think the biggest problem I faced was the software architecture. This is a hard work which needs plenty of experiences and systemic theory support. After a few of refactor and many modification iteratively, framework become more and more powerful and flexible.