Extrinsic and intrinsic calibration
Extrinsic calibration
Extrinsic calibration determines the precise position and orientation of each sensor.
Intrinsic calibration
Intrinsic sensor calibration is the process of determining how the sensor data should be modified to most accurately measure the real world.
For a camera, intrinsic calibration determines the distortion of the image caused by the lens. Lidar intrinsics, like the angle between each individual beam in a swept lidar, let you interpret the lidar’s output as a correct 3D point cloud
Intrinsic vs. extrinsic calibration
| Intrinsic | Extrinsic | |
|---|---|---|
| What it describes | Properties internal to the sensor | Position and orientation of the sensor in space |
| Camera example | Focal length, principal point, lens distortion | Transform from camera frame to robot base frame |
| Lidar example | Beam angle offsets | Transform from lidar frame to robot base frame |
| IMU example | Scale, bias, axis alignment | Transform from IMU frame to robot base frame |
| Output format | Model-specific parameters (focal length, distortion coefficients) | 6DoF pose: translation [x, y, z] + rotation (quaternion or matrix) |
Accurate multi-sensor fusion requires extrinsic calibration, intrinsic calibration, and time offset information.
Why extrinsic calibration accuracy matters
Small errors in extrinsics propagate into every perception algorithm that uses sensor data. A 1 cm translation error or 0.1° rotation error in a lidar-camera extrinsic causes visible misalignment between projected lidar points and camera pixels. This error is enough to degrade object detection, mapping, and localization.
The effect scales with distance: a 0.5° rotation error at a 20-meter range produces 17 cm of positional error. At highway speeds, or in a tightly packed warehouse, that margin matters.