Getting Realsense2 to work on Raspberry Pi3

[vrms]

https://github.com/IntelRealSense/librealsense/blob/master/doc/RaspberryPi3.md

the above page describes installing Realsense2 on Raspberry Pi3 but it does not clearly explain the proper steps. It vaguely refers to another generalized Linux distro installation procedure which itself is confusing with so many flavors of Linux, different distros & versions etc.

I followed all that described in the above page on Rpi3 but it does n't work on RPi3.

Note that it instructs to upgrade MATE to latest version of ubuntu package whereas if you use MATE on Rpi, Ubuntu strongly advises against upgrading MATE on Rpi (only updates recommended)

It would be nice if a clear step-by-step installation procedure is included for RPi3 on the above mentioned page which will help us to use Realsense on Rpi3 with out issues (I use the D415)...

What good such self-contained Realsense modulewith a D4 could be used for, if it can not work with sophisticated, highly resourceful embedded boards such as Rpi3? ..Thanks..

[MartyG]

If your Pi can connect to the internet, a more straightforward way to ​install Librealsense on Ubuntu is a method called libuvc-backend

https://github.com/IntelRealSense/librealsense/blob/master/doc/libuvc_installation.md

[vrms]

Yes, my Rpi3 does have internet connection..

I tried the above procedure on my Rpi3 with latest Buster, MATE as well as Stretch raspian images...nothing works..

it's so surprising to see that after so many years of being in the market, still not many have been able to get these realsense modules work on these embedded platforms..

I could only get it to work on my Windows 10 & Ubuntu 16.04 LTS desktops but I don't need that as the products I have in mind need to work on the field..such as in robots & drones etc

I want to use realsense on my stand alone, headless embedded boards such as Raspberry Pi3, OrangePi, Ultra96 etc..

I have all these boards and few others but sadly none of them I could get to work with realsense as basic port of realsense on these embedded platform is lacking from Intel.

With all it's claimed D4 processing and offloading of stereo depth processing loads from the host cpu etc it does n't really live up to it's promise if it can not work on such platforms.

It would be great if Intel would put some efforts to come out with a proven cross-compilation and installation procedure for arm64 etc machines and that way there would be plenty of applications for people to use your modules on various robotic and drone applications in addition to various other uses.

I also see few discussions on the Ultra96 but that also also leads to nowhere..

https://github.com/IntelRealSense/librealsense/issues/3040

[MartyG]

Apologies for the delay in responding further.

The 400 Series was designed with industrial use in mind and is extremely flexible in regard to its hardware requirements. Pi boards do seem to have a higher than average chance of encountering issues with it though.

If you are building a headless setup, you can set a CMake customisation flag called BUILD_GRAPHICAL_EXAMPLES to False so that graphical programs that use OpenGL are not included in the build.

https://github.com/IntelRealSense/librealsense/wiki/Build-Configuration

[vrms]

MartyG, appreciate your responses and help in resolving these issues.

Yes, after disabling Graphical example builds and increasing Swap files etc sufficiently, to give more resources for compilation on Rpi3, I was able to build few very basic headless examples such as depth, distance etc. Still, when I include Graphical examples build it crashes even though it had all the dependencies in place and was even able to build all gl-libs etc.

At least, this gives me some room to focus on building those headless applications that I wanted to use this module for.