A few things I noticed:
- The cameras are too far from the object of interest and thus capture mostly the background. The object of interest should ideally cover at least 60% if not more of the image.
- You should try and put either white or black cloth or plates around your camera rig to eliminate the background and sort of force RC to detect more features within the area of interest.
- If feasible, use projectors to project random patterns onto the object of interest. Take two sets of photos. one with projected pattern for alignment, one without for texturing. Your camera modules may or may not be up to that task.
- Your cameras have a relatively low resolution (5 MPx) and they also seem to generate quite a bit of noise, which won’t let you capture the level of detail you probably want.