LOAM-Livox: mapping with a Livox Horizon and open-source code

Mechanical LiDARs give a repeatable point cloud from one frame to the next, so standard methods such as ICP and NDT work. The Horizon scans with a non-repetitive pattern, so the shape of each frame changes and matching single points is unreliable. The algorithm has to be adapted.

• Edge and flat feature points are extracted from each single frame, which cuts the data volume and speeds up the computation.
• Frames are registered with point-to-line and point-to-plane matching in scan-to-map mode, estimating each frame's position and orientation in the world coordinate system.
• The result is real-time laser odometry and a 3D point cloud map. The code is published as loam_livox and Livox-Horizon-LOAM.

• The Horizon outputs about 24,000 points every 0.1 seconds, with a field of view of 81.7° by 25.1°.
• Livox says feature extraction improves mapping accuracy and stability and speeds up the computation, but the article gives no accuracy or speed measurements.

One Livox Horizon on a mobile platform, running on Ubuntu 16.04 with ROS Kinetic, PCL 1.7, Ceres 1.13.0, Eigen 3.3.7 and OpenCV 3.3.1. The Horizon is an earlier Livox model; for current sensors see our Livox LiDAR range.

• Calibrating several LiDARs to each other, by hand or automatically
• LIO-Livox: keeping SLAM stable at highway speed and in tunnels