R200 vs ZR300 vs D400 - 3D reconstruction using Intel Aero drone

[mathsten]

Hi!

At our university we have several Intel Aero Drones equipped with the realsense r200 camera. We also have a couple of the realsense zr300 model.

An idea of an assignment is to see the feasibility of using these drones and r200/zr300 cameras to do a 3d reconstruction of a laboratory. I played with the rtabmap_ros package some time, and get some exciting results with the zr300 camera. But as I am relatively new to 3d reconstruction and to the rtabmap library, maybe you have some insight to some of my questions:

Which sensor would be best at 3d reconstruction, r200, zr300 or the new d400-series?
I read the demo about a double kinect setup, would this be possible with r200/zr300 cameras?

Do you have some other insights regarding using rtabmap with Intel aero drone for the purpose of doing 3d reconstruction?

Best regards, Mathias

[matlabbe]

Hi Mathias,

D435 series look better (in specs) than the previous versions. D435 can be also hardware synchronized if more than one is used at the same time. It is possible to use multiple r200 or zr300, but they won't be synchronized.

For indoor 3D reconstruction, I still prefer TOF (time-of-flight) sensors, though some structured light depth sensors can be good too. The realsense technology is stereo IR, which I find more noisy than previous mentioned depth sensors. They may have increased disparity quality with latest sensor versions. The motion estimation can be still good for realsense sensors, even if 3D reconstruction is not as accurate or dense than depth sensors.

The results with a drone would be similar to do hand-held scanning with this kind of camera. The difference is if the mapping is done onboard, then maybe some parameter tuning could be required on rtabmap side to reduce memory usage on computation load. If you can install ROS on it, it can also simplify rtabmap installation and usage.

cheers,
Mathieu