Intelligent Vehicle Radar Calibration Multi-Sensor Fusion Automotive Calibration Room

Calibration of autonomous vehicles is a crucial step in autonomous driving and robotics technologies. Simply put, calibration establishes the geometric and mathematical relationships between the various sensors of the autonomous vehicle, as well as between the sensors and the vehicle's coordinate system. Inaccurate calibration will cause the vehicle to misjudge or veer off course, leading to navigation failures or even collisions.

Autonomous vehicle calibration mainly falls into three categories: sensor intrinsic parameter calibration, sensor extrinsic parameter calibration (joint calibration), and chassis/kinematic calibration.

1. Sensor Intrinsic Parameter Calibration

Intrinsic parameter calibration addresses the accuracy of the sensor itself and is typically performed independently at the factory or before installation.

① Purpose of Camera Intrinsic Parameter Calibration: To obtain the camera's focal length, principal point, and distortion parameters (to eliminate barrel and pincushion distortion from fisheye or ordinary lenses).

Common Method: Zhang Zhengyou Calibration Method (using a checkerboard or dot calibration board, taking multiple photos from different angles).

② Purpose of IMU (Inertial Measurement Unit) Intrinsic Parameter Calibration: To determine the bias, scaling factor, and noise of the accelerometer and gyroscope.

Common Method: Static Allen variance analysis, multi-position rotation testing.

2. Sensor Extrinsic Parameter Calibration

Extrinsic parameter calibration addresses the positional relationships between sensors and between sensors and the vehicle body (spatial relative rotational moment R and translation vector T), which is the foundation of multi-sensor fusion.

① Camera and LiDAR

Objective: To align the 3D point cloud from the LiDAR with the pixel dimensions of the 2D image from the camera, achieving "coloring the point cloud" or "adding depth to the image."

Method: Target-based calibration: A calibration board is placed in the scene, and checkerboard corner points in the image and edge planes in the point cloud are extracted. Extrinsic parameters are solved using optimization algorithms.

Targetless (online) calibration: Edges and building lines in the natural environment are used to match the image gradient with the edge intensity of the point cloud.

② LiDAR and Inertial Navigation/Vehicle Body

Objective: To convert the data measured by the LiDAR to the center of the vehicle (usually the rear axle center or IMU center) to ensure that mapping and localization are not drifted.

Method: The vehicle performs figure-eight circling, straight-line driving, or rotating in place within the horizon or a specific terrain. The relative pose is calculated by matching point cloud features with the integral trajectory of the IMU.

③ Multi-LiDAR Calibration

Purpose: When the vehicle has multiple LiDARs (e.g., one main LiDAR + two blind spot LiDARs), stitch their point clouds together to form a complete 360° view.

Method: Extract common features (e.g., ground, walls) from overlapping areas and use the ICP (Iterative Closest Point) algorithm for registration.

3. Chassis and Kinematic Calibration

Chassis calibration addresses the accuracy of the actuators and wheel speedometers, preventing the vehicle from "thinking it has traveled one meter when it has only traveled half a meter."

Wheel Diameter and Track Calibration: Calibrate the actual effective radius r of the two drive wheels and the actual distance W between the wheels on both sides.

Odometer and IMU/LiDAR Extrinsic Parameters

Method: Have the vehicle travel a straight distance or rotate 360° in place, compare the data recorded by the wheel speedometer with the absolute displacement and angle measured by the LiDAR/INS, and correct the wheel speedometer's scaling factor.