Real-Time RGB-D Mapping and 3-D Modeling on the GPU Using the Random Ball Cover

摘要: In this chapter, we present a system for real-time point cloud mapping and scene reconstruction based on an efficient implementation of the iterative closest (ICP) algorithm graphics processing unit (GPU). Compared to state-of-the-art approaches that achieve performance using projective data association schemes which operate 3-D geometry solely, our method allows incorporate additional complementary information guide registration process. work, ICP’s nearest neighbor search evaluates both geometric photometric in direct manner, achieving robust mappings real-time. order overcome bottleneck space traversal, exploit inherent computation parallelism GPUs. particular, have adapted random ball cover (RBC) structure algorithm, originally proposed high-dimensional problems, low-dimensional RGB-D data. The is validated object scenarios. Our achieves frame-to-frame runtimes less than 20 ms off-the-shelf consumer GPU.