Visual SLAM on Android Device using single-frame Pointclouds

Question

I have developed a Python application that uses the Intel RealSense D455 camera and the library open3d to perform realtime and, later, slower high definition 3d reconstruction.
The algorithm used to achieve this is SLAM.

Now, I'm looking everywhere to find a library for Android that can do SLAM or Visual Odometry using data coming from an external RGB-D camera, but I can't seem to even find a commercial SDK that does that. ARCore would be cool, but it can't use any external camera stream, Open3D seems to have a project that compiles it (only through version 0.9), but I can't get it to work.

What I actually need is only the merging of the single-frame pointclouds, so ICP registration may also be helpful.

Anyone has any idea what could help?
Maybe Unity with some plugin?

As far as i know, ARCore MUST use the internal camera of the device. What I need is something that can use data from an external device (either the RGB and Depth images or pointcloud calculated from those) — vittoema96, Apr 14 '23 at 17:52
so you're on android, but you don't want to use the device camera? — Christoph Rackwitz, Apr 14 '23 at 18:56
no, I want to use an Intel RealSense D455 as input, as it has higher definition regarding depth — vittoema96, Apr 17 '23 at 13:53

score 0 · Answer 1 · answered May 05 '23 at 14:21

Ok, for anyone that has my same problem, I found this library that seems to do something of what i need. It's really really hard to use tho, much harder than open3d, i'm still trying to make something out of it. The library is BoofCV, it is android compatible and writable in Java (instead of native C++) I'm still resolving some OutOfMemory errors, but it seems promising.

Still looking for a final solution, so if anyone has a better way to do this, please tell me.

Visual SLAM on Android Device using single-frame Pointclouds

1 Answers1