In the geometric calibration, we use the three cameras and one ABB robot arm setup to conduct our experiments(Fig 7.). We would use the images that captured by the camera-robot system to conduct the geometric calibration. We wanted to test two things, one is we could get the high-quality images from path planning part, and the other is our algorithm could deal with these image in an efficient method and provide the good accuracy.
Camera and robot arm setup
performance evaluation
After the image acquisition, we got 80 raw images per camera to conduct the camera calibration. The process time of the camera calibration was quite efficient and the RMSE reprojection error was 0.1698 pixel(Fig 7.) which was meet our SVE requirement that the reprojection error should less than 0.2 pixels
Geometric calibration result( left: reconstruction result; right: reprojection error)
We also compare the denoised image and raw images in the geometric calibration. The result was quite similar, so in the geometric calibration, we could directly use the raw image and saved time doing denoising.