Alexandre, an internship student from UPSSITECH/SRI has been hired in CNRS-LAAS. From 2020 April to July, he worked on the calibration of the perception system embedded in the Hunter 2.0 robot. This system is made of four realsense cameras and one 2D laser. The objective was to express the point clouds provided by all the cameras with respect to one common frame attached to the 2D laser device. Different algorithms have been tested. Finally, the obtained results have shown that [1] was the best suited. Several achievements have been made during this internship:
- Development & validation using ROS of the chosen calibration algorithm
- Development of interfaces to ease the calibration process: visualization of the image provides by each camera and the 2D laser, selection of the points to be matched for the calibration process, etc.
The results are presented here below. The first figure shows an example of an erroneous calibration process: the point clouds given by two cameras are projected in the 2D laser frame, and do not match. The second figure is obtained when the calibration is performed using the chosen algorithm. It can be seen that all point clouds match quite well. Our tests have also shown that using about 10 images is sufficient to get an accurate calibration result.
- Two camera point clouds projected in the 2D laser frame without proper calibration:
- Both point clouds, once the chosen calibration algorithm is applied:
