Robotics StackExchange | Archived questions

does rtabmap actually use computing power of jetson tx2 with opencv3.3?

I am asking about proper installation of rtabmap and its components on nvidia jetson tx2, to fully use computing power of this board.

I read from rtabmap_ros github page:

To use rtabmapros on Jetson, you can follow the instructions above if you don't care if OpenCV is built for Tegra. However, if you want rtabmap to use OpenCV 4 Tegra, we must re-build visionopencv stack from source too to avoid conflicts with vision_opencv stack binaries from ros (which are linked on a not optimized version of OpenCV). Here are the steps:

...

1) Does it means that rtabmap works properly on jetson only with opencv4?

2) (not related to rtabmap) In nvidia sdkmanager I see that to install opencv4 I need jetpack 4.3 on ubuntu 18.04. Is there any possibility to install opencv4 on ubuntu 16.04 with jetpack 3.3, L4T 28.2.1? if yes, will rtabmap run properly in this case?

Asked by june2473 on 2020-02-12 01:49:13 UTC

Comments

Answers

That post is relatively old, when OpenCV4 was not released yet. "OpenCV 4 Tegra" here means "OpenCV FOR Tegra", which was the term used by NVidia at that time. This walkthrough was tested on 16.04 Jetpack 3, not sure it is still working for jetpack >=4 (18.04). In your case it seems what you want.

Normally, rtabmap should work on Jetpack 3 (16.04) and Jetpack 4 (18.04). Afterwards, it is more if latest opencv4 version is still using tegra optimization on older jetpacks.

Asked by matlabbe on 2020-02-20 18:53:55 UTC

Comments