Real-world 3D data is locally dense but globally sparse. Therefore, efficient sparse data structures are an essential component of dense 3D perception for computer vision and robotics. We manifest the power of spatial hashing by two typical tasks: dense scene reconstruction and global registration. In the first task, we accelerate volumetric integration and surface reconstruction by introducing a GPU-based spatial hash table with specific optimizations. For global registration,
we improve the registration accuracy of point cloud pairs via sparse convolution networks based on spatial hashing. Upon these, we propose a unified spatial hash map that is 1) aimed at general for tasks from classical reconstruction to deep spatially varying perception; 2) optimized on GPU with vectorized operations; 3) user-friendly and
compatible for both GPU and CPU.
Michael Kaess (Advisor)
Zoom Participation. See announcement.