Support Website Contact Support Blog

Cameras uncalibrated

Help please…been using Pix4d to create textured 3D mesh of simulated coronary artery anastomoses. The original images were created using a turntable and Canon 600D on a tripod. Three sets of approx 30 pics at 10-15deg angles each covering 360deg at different elevations, turntable up and down, with infinity background. I have used Photoscan to create a reasonable quality 3D mesh but was relying on pix4d to allow me to create a video animation trajectory, to give trainees surgeons and their supervisors a concept of flying around the anastomosis as they assess it externally. I have imported 90 pics and generated a point cloud and textured mesh and been able to create a video animation of good quality. Just what I needed. Only half the cameras are calibrated so the backside of the textured mesh is poor quality. I have tried double image size on Keypoint Image Scale which improved the number of calibrated cameras but textured mesh still not good enough and only 270deg of cameras calibrated. I have also tried using only one 360deg set of pics at and this had the same effect cameras from only 180 degrees were calibrated, the textured mesh was even worse.
Do you have any suggestions, I have tried to import the point cloud saved from Photoscan as a .las file but this does not seem to work. I also tried to recalibrate the uncalibrated cameras but ultimately would rather fix the problem as this seems very time consuming.
Very grateful for any ideas, attached reports and screenshots of textured mesh and camera positions
Thanks, James

Dear James,

One explanation for this is that the background is different on the two sides and this makes it harder for the software to recognize the objects. In any case, it turns out that many of your cameras are uncalibrated and that is why you get these results. It is possible to calibrate those cameras by following this procedure

This could be caused by processing the project with the fast template instead of the regular one. It runs faster by extracting less features and the quality of the reconstruction might suffer. 

Best regards,