RTAB map VS Photogrammetry

[sachet]

Hi,

I am new to this topic and I am sorry if this question doesnt belong here.

I wanted to know which method is better suited to get accurate pointclouds like laser scanners (such as leica). Particularly meshroom vs rtab map?

I could not find any literature comparing dense monocular slam and  photogrammetry.

I would be thankful if you could provide me with any resources.

Thanks.

[matlabbe]

Hi,

RTAB-Map is not a monocular SLAM approach, it does RGB-D, Stereo or LiDAR SLAM.

Given an area/object with a lot of discriminative textures and high-res imagery, photogrammetry will give pretty close result to LiDAR, up to a scale factor. The process is often done offline, taking pictures approximately and hoping that photogrammetry approach will find enough correspondences. With experience, you can know how to take good pictures to improve results.

SLAM approaches are more often online, you can get a live feedback of the scanning. It is easier to see what is scanned so far. If you scan indoor, it may also be easier to get better results with a RGB-D camera or a LiDAR (which RTAB-Map supports) to "see" textureless surfaces like white walls.

You may search keywords like Photogrammetry, structure from motion and visual SLAM comparison.

cheers,
Mathieu

[sachet]

Hi Mathieu,

Thank you so much for the response. I had one morr question. Could the output of SLAM i.e camera poses be used as input for photogrammetry for better results compared to a total SLAM or total photogrammetry system?

In short, if I have a monocular camera, what is the best method for me to get a dense pointcloud of the environment?

Thanks again.

[matlabbe]

Hi, it could be possible to use a Monocular SLAM approach (like ORB_SLAM or VINS) in order to get poses of the images. Feeding poses to a photogrammetry system may help it to converge faster. Things to check:
1) The monocular SLAM approach should be able to export poses to a file,
2) The poses may not be in the same coordinate frames used by the photogrammetry system, so poses may need to be transformed in photogrammetry coordinate frames,
3) Can photogrammetry system actually import poses and in which format?

In short, if I have a monocular camera, what is the best method for me to get a dense pointcloud of the environment?

Photogrammetry software (assuming you have high resolution images and no limited time to process all those images).