extrinsic calibration

Published 2020-11-21

Problem motivation and description:-

The problem of interest here is to find the pose of a camera with respect to its base frame in a completely online fashion. The primary motivation behind this problem is that motor-impaired wheelchair users often crash into walls during operation, thereby damaging the pre-calibrated camera pose. As a first step towards solving the problem, I constructed an RGBD visual odometry using a Realsense camera. By measuring the camera ego-motion (A) through visual odometry and the base frame ego-motion (B) through wheel odometry between two synchronized time stamps, I was able to optimize for the rigid body transformation X between the camera and the base frame in the matrix expression AX = XB using Gauss Helmholtz optimizer. But I realized that only 5 out of the 6 rigid body transformation parameters are observable due to the planar motion of the robot. Therefore to estimate the 6th parameter (Z translation), I calibrated the ground plane and estimated the height of the camera above the ground plane after I realized that this height is indeed the otherwise unobservable Z translation parameter.

Video description:-
In the video, you will see the wheelchair moving through the simulation environment in Gazebo. As the wheelchair moves, it accumulates measurements of ego camera motion and ego base frame movement. The measurement count at any instant is shown in one of the tmux split screens. Once the number of measurements reaches 1000 measurements, the Gauss Helmholtz optimizer is triggered and an estimate for the transformation between camera and base frame is obtained.