Hi everyone,
I am working on an outdoor RF propagation measurement campaign using a Blackmore FMCW Doppler LiDAR, which captures point clouds with an additional per-point radial velocity channel alongside the standard XYZ coordinates. My goal is to generate a high-quality mesh of the building facades and surrounding environment for use in RF ray tracing simulations — meaning I need geometrically clean, flat surfaces rather than photorealistic detail. I have a dense fused .ply point cloud captured from multiple LiDAR positions and orientations around the scene, and I am currently using Open3D and Python for processing and Blender for visualization and final mesh preparation.
My main challenges are: (1) removing noise while preserving sharp building (or object) edges, (2) generating flat wall and roof surfaces without holes or artifacts, and (3) understanding whether a multi-position scanning setup gives me any advantage in mesh quality. Currently, I am capturing the scene with the LiDAR at a fixed location and orientation and scanning a static environment. So capturing for an extended amount of time results in noisy point cloud. So far, I tried statistically filtering the outliers, then Poisson and BPA as meshing methods. The surfaces look rough and the edges lose lots of information.
Any advice on recommended algorithms — such as Poisson reconstruction, RANSAC plane fitting, or alpha shapes — or workflow tips for this use case would be greatly appreciated! Also any advice on the capturing setup would be appreciated!
I am working on an outdoor RF propagation measurement campaign using a Blackmore FMCW Doppler LiDAR, which captures point clouds with an additional per-point radial velocity channel alongside the standard XYZ coordinates. My goal is to generate a high-quality mesh of the building facades and surrounding environment for use in RF ray tracing simulations — meaning I need geometrically clean, flat surfaces rather than photorealistic detail. I have a dense fused .ply point cloud captured from multiple LiDAR positions and orientations around the scene, and I am currently using Open3D and Python for processing and Blender for visualization and final mesh preparation.
My main challenges are: (1) removing noise while preserving sharp building (or object) edges, (2) generating flat wall and roof surfaces without holes or artifacts, and (3) understanding whether a multi-position scanning setup gives me any advantage in mesh quality. Currently, I am capturing the scene with the LiDAR at a fixed location and orientation and scanning a static environment. So capturing for an extended amount of time results in noisy point cloud. So far, I tried statistically filtering the outliers, then Poisson and BPA as meshing methods. The surfaces look rough and the edges lose lots of information.
Any advice on recommended algorithms — such as Poisson reconstruction, RANSAC plane fitting, or alpha shapes — or workflow tips for this use case would be greatly appreciated! Also any advice on the capturing setup would be appreciated!
