cartographer/sensor/internal/voxel_filter.h

Overview

VoxelFilter downsamples point clouds by dividing space into 3D voxels and keeping one point per voxel (typically the first or centroid).

Class: cartographer::sensor::VoxelFilter

Construction

explicit VoxelFilter(float voxel_size);

  • voxel_size — Voxel edge length in meters

Filtering

PointCloud Filter(const PointCloud& point_cloud) const;

Returns downsampled point cloud.

Algorithm

  1. Compute voxel index for each point: index = floor(point / voxel_size)

  2. Use hash map to track first point per voxel

  3. Return collected points

Complexity

  • Time: O(N) for N points

  • Space: O(M) for M occupied voxels

Usage

// Create filter
sensor::VoxelFilter filter(0.025f);  // 2.5cm voxels

// Filter point cloud
sensor::PointCloud raw = lidar->GetScan();
sensor::PointCloud filtered = filter.Filter(raw);

// Typical reduction: 100k points -> 10-20k points

Configuration

Used in trajectory builder configs:

TRAJECTORY_BUILDER_2D = {
  voxel_filter_size = 0.025,  -- 2.5cm
}

TRAJECTORY_BUILDER_3D = {
  voxel_filter_size = 0.15,   -- 15cm (3D lidars denser)
}

See Also