A Method for Enhancing Robustness of Viewpoint Planning Based on Spatial Supervoxels

In the viewpoint planning task for unknown objects, the accuracy of the perception system and calibration lead to the collected point cloud data containing numerous noise and stratification, which interferes with the viewpoint planning process.To address the above problems, this paper proposes a method to enhance the robustness of viewpoint planning based on spatial supervoxel.First, the spatial voxel environment is constructed to store the triangular meshed point cloud containing noise, which is used to record the positional distribution information of the point cloud in space; then, the voxels containing the point cloud are neighborhood-grown to form a Supervoxel unit, which expands the receptive field; And finally, the point cloud boundaries are detected and categorized using spatial Supervoxel, which are used to guide viewpoint planning.To verify the effectiveness of the proposed method, this paper develops a robot perception system, uses an aircraft radome as the experimental object and introduces varying levels of system errors for the viewpoint planning experiment.The experimental results indicate that the proposed method can accurately detect and classify point cloud boundaries to guide viewpoint planning, achieving complete surface reconstruction of unknown objects, even in the presence of a certain degree of noise in the perception system. The overall modeling time is approximately 7 minutes, demonstrating strong robustness and real-time performance.