LiDAR Camera Calibrator

Manually refine static LiDAR-to-camera calibration with the data you already have.

LiDAR Camera Calibrator is an open-source PySide6 package for manually refining static LiDAR-to-camera calibration. It provides synchronized playback, a navigable 3D point-cloud view, projected camera overlays, arbitrary-camera calibration, and direction-explicit JSON results.

It is intended for the practical work after a vehicle rig has been assembled: tightening alignment for labeling, validating a sensor setup, investigating a new calibration hypothesis, or experimenting with a different projection. Because vehicle rigs often remain stable for long periods, a deliberate manual refinement pass can be useful when an automatic calibration is unavailable or does not meet the precision needed for a workflow.

Use any dataset

The package accepts semantic sensor data, not a particular vehicle dataset or directory layout. Your data can come from raw/proprietary files, Python and NumPy arrays, a custom lazy loader, or a supported standard Foxglove MCAP recording. KITTI is available as an optional, tested example pipeline.

The generic integration boundary is:

• timestamped LiDAR point clouds;
• one or more timestamped camera streams;
• camera intrinsics and image dimensions;
• a connected, acyclic transform path from every sensor to IMU;
• timestamped world-from-IMU poses covering the LiDAR timeline.

Provide those values through ordinary Python/NumPy sequences, lazy application-owned sequences, or the included Foxglove adapter. The profile writer validates and synchronizes the complete dataset before creating portable lidar-camera-scene/1 MCAP.

Workflow

from lidar_camera_calibrator import CalibrationConfig, launch_calibrator, write_profile_mcap

write_profile_mcap(source_config, "scene.mcap")
result = launch_calibrator("scene.mcap", CalibrationConfig())

The separation is intentional:

1. Your source integration interprets raw or standard source data.
2. The profile writer validates and atomically normalizes it.
3. The canonical reader/viewer operates on one stable contract.
4. Calibration JSON remains separate from immutable scene recordings.

Choose an integration path

Your data	Start here
NumPy arrays or Python objects	Direct data integration
Proprietary/raw files	Raw files and custom loaders
Foxglove JSON/base64 MCAP	Foxglove MCAP
KITTI test data	KITTI development adapter
Existing canonical scene	Launch the viewer

Stability

The package is currently alpha (0.1.x). The serialized lidar-camera-scene/1 schemas are frozen, while Python APIs follow semantic versioning as the first public release is prepared.