Can you use intel realsense + photogrammetry together?
I would like to know
Can i use high resolution cameras along side intel realsense depth cameras to generate depth maps for high resolution wrinkle details in face scans? Can photogrammetry be used in ADDITION to depth cameras to achieve a higher resolution result?
Also, can multiple overlapping realsense cameras be used to improve the quality and resolution of the scans past what a single camera is capable of? Does this functionality exist within the SDK?
Is the SR305 sufficient for facial motion capture / facial scans given its short range specialty or is the D400 series necessary for this kind of work?
Also, besides facial scanning i have plans to use this for facial motion capture. one of the ways it has been traditionally done is with retroreflective markers like this:
http://www.roborealm.com/tutorial/FIRST/slide010.php
in this example they are using the kinect and diffusing the ir projector so only the retroreflective markers will show up. obviously this gives a result that is not the usual kind of information used by the SDK but achieves a much much higher accuracy result for motion capture solutions, not suited for depth detection but is perfect for motion tracking. for facial performance capture, i can place these markers on key areas of the face and diffuse the IR projector, so that the face would not be visible to the depth camera, everything would be black except for the tracking markers. would stereo IR solutions work well for this atypical usage?
-
1. Yes, you can sync RealSense cameras with non-RealSense cameras. An example is combining RealSense depth data with high-definition RGB from a non-RealSense camera. The two links below describe ways to do it.
https://github.com/IntelRealSense/librealsense/issues/2186
https://github.com/IntelRealSense/librealsense/issues/4202
2. You can overlap RealSense 400 Series depth cameras to improve the quality of depth results. This is because in the areas where the fields of view overlap, there is redundancy in the data caused by more than one camera covering the same area. 400 Series cameras do not interfere with each other when in close proximity. SR300 / SR305 cameras do experience interference when in close proximity to each other though.
If you are seeking to generate a solid 3D model from multiple RealSense cameras, the professional version of RealSense-compatible 3D scan software called RecFusion can do this. RecFusion has been highly recommended by RealSense 400 Series camera users who have used it.
https://www.recfusion.net/index.php/en/features
3. Whilst the SR305 is able to be used for 3D model scanning, the SR300 technology inside it is a couple of years older than the 400 Series cameras and so you will be able to achieve a higher resolution with the 400 Series cameras. The 400 Series is the best solution in the RealSense range for 3D scanning of the human body.
Of the 400 Series models available, the D415 model offers the best quality image, as it has around 2.5x the image quality of the D435 model. The D435 model excels in tracking very fast motion such as vehicles, but would not offer a great advantage over the D415 in this application (scanning of slow-moving human bodies). Even though the D415 has a slower shutter than the D435 that makes it unsuitable for following fast motion without blurring / motion artifacts on the image, it can cope fine with human walking pace.
A company called Jaunt did 360 degree avatar capture with what appeared to be the D415 model, with six cameras used.
Intel also did a seminar for advanced users in creating markerless body tracking with RealSense and multiple cameras.
https://www.youtube.com/watch?v=VSHDyUXSNqY
Intel have invested heavily in markerless capture as the future of data-focused volumetric video production, such as their huge capture dome facility at a studio near LAX airport.
Please sign in to leave a comment.
Comments
1 comment