Module icp 

GPU-accelerated ICP

The target cloud is uploaded once as a kd-tree. Each iteration runs one kernel that moves every source point by the current estimate, finds its nearest target point in the tree (starting from last iteration’s match), and sums the values the update needs within each workgroup. Only those small per-workgroup sums come back to the CPU.

Structs

BatchICPJob

Batch ICP operation for multiple point cloud pairs

BatchICPResult

Result of a batch ICP operation

GpuPointToPlaneICPResult

Result of GPU point-to-plane ICP

Functions

gpu_batch_icp

Execute batch ICP operations on multiple point cloud pairs

gpu_icp

GPU-accelerated ICP registration

gpu_icp_point_to_plane

GPU-accelerated point-to-plane ICP registration.