Installing OpenCV is Like...

Question

After several days into attempting to install OpenCV on my BB-AI it feels like I'm at about 0:15s into this video !

https://www.youtube.com/watch?v=wmFzGXy5uYc

Maybe some pointers for others are in my iBox posts:

I also tried cross compiling on my faster PC with the aim of porting the code across afterwards, but that too failed with the libpython2-dev:armhf library not being found. After much adjustment of my machine on the Linux command line I'm also no nearer on that approach.

But it did make me think, hence this question rather than a blog post:

If I were to compile the OpenCV on my other machine and port it across to the BB-AI then has someone else done this? e.g. why am I sitting waiting days for the BB-AI to run CMAKE if it could be done externally and highly likely someone already has.

I've learnt many Linux commands along the way and have a rough understanding of what cross-compiling is.....so progress todate has been worthwhile still

shabaz · Accepted Answer

Hi Rod!

I can't recall how long it took to compile OpenCV from scratch on a BeagleBone Black, but it's doable (within hours but cannot remember how many!), so BB-AI should work nicely too.

If I can make a suggestion, it would be to avoid the cross-compiling for now, and do it all directly on the BB-AI. The reason is, cross-compiling is delicate, any slight difference can result in incompatibility of the executable, or absence or incompatibility of dynamically-loaded code.

Cross-compiling is useful for some use-cases, but just for speed improvements in build time is not so important unless you're doing commercial development (and even then, it's normal for builds to run overnight - not that you need to do that for code the size of OpenCV!).

A major project I spent several years on (a telco solution), I had to do cross-compiling, until, over the years, the processor speed of the target hardware, (and resources on the target) significantly improved, so that I didn't need cross-compiling speed boost any more. I also benefited from not needing to maintain the cross-compile environment any longer, which was an overhead for a small team (me and my team-leader).

(I believe the Pi people do the same thing, although I could be wrong - I think I saw some post or comment ages ago that their build server is a bunch of Pi's, and not cross-compiled, although they may have originally cross-compiled in the Pi 1 days).

There could also be other reasons to continue cross-compiling in work environments (e.g. if you cannot get your target hardware approved for installation in the data centre, infosec approval etc) but that won't apply to you.

Finally, there are cross-compiling benefits where additional PC software tools are being used, or better PC tools are available, but again these don't apply here.

If you're running low on storage then it is worth mounting some extra storage to the BB-AI. I sometimes use my NAS for that. It's not fast, but if you're only compiling OpenCV rarely, then it's not a bad way to go.

14rhb · Answer

Hi Shabaz,

Thank you for that good advice and insight into what cross-compiling is about.

I'm back on the BB-AI directly, albeit via SSH, and having another go at following the instructions on the OpenCV website https://docs.opencv.org/4.1.1/d7/d9f/tutorial_linux_install.html

When you mention storage, the BB-AI comes with 1GByte of onboard RAM so I guess you mean adding an SDCard to augment that? I'll have to read about that as my Linux skills are getting pushed on this project (in a good way). I have a 4GByte uSDcard free.

I also have the HDMI plugged in and see there is a Debian desktop up and running. I could probably turn that off somehow and save RAM.

shabaz · Answer

Hi Rod,

The 1GByte RAM should be fine (as you say, not running the desktop should speed things up) since the older BBB can compile OpenCV too. Regarding storage, if you've got any existing home NAS (e.g. to store movies/photos etc) then sometimes there is an option in there to turn on NFS (network file sharing) and then a mount command (I can dig up the syntax if needed) can be used in Linux so that you can go to (say) /mnt/nas/opencv and put all your files there (this is useful where there's a lot of source code) but a micro SD card should be fine too.

vlasov01 · Answer

Hi there,

I've run in out of memory situation during compilation on BB-AI. If you get the same issue, than you will need to create a swap.space. I was following the following document: https://linuxize.com/post/how-to-add-swap-space-on-debian-9/ to accomplish it..

fyaocn · Answer

That is really fun digging into the Framework and Software Structures. But only one month left for the Vision Things, I am really eager to see how your project going on. Have to admit that, it is not easy.

fyaocn · Answer

Here it is. You can #include directories from /usr/include , etc.

skywalker1211 · Answer

i don't have a BB AI. i install opencv4 into raspi2.

vlasov01 · Answer

Thank you for the reply! I'm building now OpenCV 4.1.2 on BB AI. So I'll get my answers soon.

fyaocn · Answer

in fact,you can find opencv2 on B.B. Ai already. If you really need opencv 4.1, build from sources then.

shabaz · Answer

I didn't realize it was up to v4.x already! I've only used opencv2, it has the advantage perhaps that there's lots of literature on it.. at least, the books I found at the time I tried it all referred to opencv2.

skywalker1211 · Answer

there is no error message. i got the same problem when i compile opencv4 in raspi2. increase swap space solve the problem. https://linuxize.com/post/how-to-install-opencv-on-raspberry-pi/ ---> "Start by increasing the swap space to avoid compilation hangups due to memory problems:"

vlasov01 · Answer

Hi Rod, I was running dmesg when I realized that the build process run out of memory and got killed. I don't remember exact message so. If I got one again I'll let you know. Sergey

clem57 · Answer

fyaocn wrote: in fact,you can find opencv2 on B.B. Ai already. If you really need opencv 4.1, build from sources then. So use opencv2 instead of opencv?

skywalker1211 · Answer

14rhb i changed from 100 -> 1024, and revert after installation.

vlasov01 · Answer

I'm wondering if you noticed any performance differences between supplied OpenCV 3.2 and v4 for image processing tasks? Is v4 still leveraging DSP and EVE cores of AM5729?

14rhb · Answer

This is most helpful, thank you. I naturally assumed to go for the latest version (4.1.1) and didn't think to look back. Are you saying OpenCV2 is already installed as part of the build?

14rhb · Answer

Thank you fyaocn that was a real insight and is going to help me along really well - seems like I have v3.2.2 installed by default....now to try it out.

yosoufe · Answer

What about cross-compile with docker container? i haven't done it but it may work I found a page for cross-compiling for raspberry pi in docker container here: https://desertbot.io/blog/how-to-cross-compile-for-raspberry-pi

Installing OpenCV is Like...

Top Replies