Projects: Robot Camera Calibration

As part of various projects I’ve had to implement solutions for intrinsic and extrinsic calibration, including:

  • Camera intrinsic and extrinsic parameters (focal length, distortion)
  • Camera to 2D sensor (another camera)
  • Camera to 3D sensor (depth, lidar)
  • Camera-to-Robot (mobile base, manipulator arm)

Sometimes you can assemble the robot, calibrate everything once, and then the sensor offset transforms remain constant for an extended period. However if you frequently change sensors or swap the end-effector then automatic calibration is a better solution, allowing the robot to find the parameters by itself.

This page has some selected diagrams and photos related to my work on calibration.

Fig. 1: Intrinsic calibration using a
checkerboard (chessboard pattern).
Fig. 2: Automatic calibration of camera intrinsic
parameters. The robot is programmed to
move in a sequence of configurations.
Fig. 3: Diagram illustrating the concept of camera-to-gripper calibration:
comparing sensor-in-hand (left) Vs. an external fixed sensor (right) [1].
Fig. 4: The robot grasps a piece of lumber
such that two planar faces are visible,
providing a constraint for calibration.
Fig. 5: Calibrating a Kinova Mico arm with
a depth sensor fixed to the wrist.
Fig. 6: The robot grasps an AprilTag for automatic
calibration of a downward-facing depth sensor.
Fig. 7: A sphere (ball on a stick) for calibrating
multiple depth sensors looking at a scene.
Fig. 8: Calibrating camera-to-LiDAR
using checkerboard (chessboard pattern)
and planar surface.
Fig. 9: Aligning the data from multiple sensors is the first step in a sensor-fusion
pipeline. This shows the LiDAR beams overlaid on top of a camera image [2].
[1] Diagrams by T. A. Myhre
[2] Figure adapted from:
G. A. Kumar, J-H Lee, J. Hwang, J. Park, S. H. Youn and S. Kwon, “LiDAR and Camera Fusion Approach for Object Distance Estimation in Self-Driving Vehicles”, MDPI Symmetry, February 2020.