Newbie here...Anybody successfully compile OpenCV on a rpi zero?

[steelo]

Hi everyone,

 

I spent about half the weekend pulling my hair out trying to get openCV to compile on a rpi zero. There are countless (and very tedious) guides one can find online and I followed them to a 'T'. It involves terminal and installing multiple libraries from repositories and praying that they are still accessible....I think there are about 30-40 steps altogether. I really have no idea what I'm doing really, so if anything goes wrong I'm pretty lost. The first attempt I was (kind of almost) successful but the compile failed at about 29% after running for about 3 hours. After attempting this multiple times, I decided to try to find an easier method...however, there doesn't appear to be any. I know it is possible to compile this on a zero, but I am still too green to be able to get past any errors that aren't listed in the step-by-step tutorials. Has anybody had any success for this? I'm very surprise nobody (who actually knows what they're doing) has already created a script to make this easier?

 

I'm really wanting to better learn python and eventually write code to be able to utilize a webcam, but the fact that it is this difficult just to get the software I need on there is not at all encouraging.

[tikker]

5 hours ago, steelo said:

I'm very surprise nobody (who actually knows what they're doing) has already created a script to make this easier?

Some steps require setup/input from your side or sudo and running a random script from the internet with sudo is about as bad as running totallyamovieandnotavirus.mp4.exe on Windows. Not saying you'll get a virus or destroy your system, but it's much clearer to have the steps to do written out rather than having a "run this and voila" tutorial. You'll not know what went on nor where it was installed or what it installed. Plus the RbPi community are usually tinkerers

 

What errors specifically did you run into? What guide did you follow? https://www.pyimagesearch.com/2015/12/14/installing-opencv-on-your-raspberry-pi-zero/ seems pretty straightforward.

[steelo]

8 minutes ago, tikker said:

Some steps require setup/input from your side or sudo and running a random script from the internet with sudo is about as bad as running totallyamovieandnotavirus.mp4.exe on Windows.

 

What errors specifically did you run into? What guide did you follow? https://www.pyimagesearch.com/2015/12/14/installing-opencv-on-your-raspberry-pi-zero/ seems pretty straightforward.

That's the same guide I've been using! Good pick

 

Yes, I was hoping that it would be straightforward and it is (just copy and paste the code)...the problem are the unexpected errors I'm receiving. I'm mainly having issues with the CMAKE portion. The first time I was able to run MAKE but it crashed at 29% and gave me an 'error 2' I did create a 1024MB swap file as the guide suggested. 

 

After the initial crash, I was thinking I could just run the 'make' command again, but it immediately gave an error like 'no path' or something similar. Honestly, I barely know what I'm doing...I'm trying to follow the instructions verbatim, but when things go wrong, I don't know where to go from there.

 

 

[tikker]

3 minutes ago, steelo said:

That's the same guide I've been using! Good pick

 

Yes, I was hoping that it would be straightforward and it is (just copy and paste the code)...the problem are the unexpected errors I'm receiving. I'm mainly having issues with the CMAKE portion. The first time I was able to run MAKE but it crashed at 29% and gave me an 'error 2' I did create a 1024MB swap file as the guide suggested. 

 

The second time and third time, I tried running MAKE, it immediately gave an error like 'no path' or something similar. Honestly, I barely know what I'm doing...I'm trying to follow the instructions verbatim, but when things go wrong, I don't know where to go from there.

 

 

2 is make's return code for errors. If you scroll up (or run make again immediately to have less output) you should see an actual error message somewhere. We'll need that to get a better idea where to start.

[steelo]

21 hours ago, tikker said:

2 is make's return code for errors. If you scroll up (or run make again immediately to have less output) you should see an actual error message somewhere. We'll need that to get a better idea where to start.

Thanks for the reply. I'm still very green when it comes to Linux. Unfortunately, I have since reflashed my sd card so I wouldn't be able to give you the actual error message. I think I might have found an 'easier' OpenCv' tutorial (I did take note of your warning that it may be a dangerous to go that route). The author claims to have made a script to automate everything and provided a link where you can look at the code if you want to. I started it when I was home at lunch about 2 hours ago and will have to see if it successfully completed when I get home. I can't find the link now on my work computer, but he is claiming it will take ~3 hours to install.

 

I've also just now found this...https://hackaday.io/project/7008-fly-wars-a-hackers-solution-to-world-hunger/log/23068-installing-opencv-on-a-raspberry-pi-the-easy-way

 

Would this provide everything I need to create projects in python using a webcam? I don't know why there are so many tutorials out there that take 9 hours to install and other 'easy guides' that claim to only take a few minutes to a few hours??? Are they pre-compiled? (if that's a word) What is the difference and are these 'quick' guides missing something I may need?

[BuckGup]

I suspect you are doing vision tracking on a pi. Our FRC team just cross compiled it for the pi on a different PC and simply ran it on the pi afterwards. I wish I knew more info but I just followed a guide online

[steelo]

18 hours ago, BuckGup said:

I suspect you are doing vision tracking on a pi. Our FRC team just cross compiled it for the pi on a different PC and simply ran it on the pi afterwards. I wish I knew more info but I just followed a guide online

Yes, that is correct....I was FINALLY able to get it to work last night on my pi zero using a simplified tutorial. https://hackaday.io/project/7008-fly-wars-a-hackers-solution-to-world-hunger/log/23068-installing-opencv-on-a-raspberry-pi-the-easy-way  I recall reading over and over about how important it is to create a 'virtual' environment to work in (cv2) I don't exactly understand why and don't know if that's even an option with how I have it set up currently.

 

I played around with some of the demo python projects and I came to the realization that the rpi zero does not have the processing power to effectively track an object. The one demo I ran drew a line as it tracked a yellow highlighter but there was a good 5-10 second delay in the tracking. I have a rpi 3 in the mail, which I hope will make it easier to set up OpenCV and run it much more smoothly. On another note, for some reason, I'm having to execute these .py demos in terminal - If I try running from within Thonny, I'll receive numerous errors on the bottom pane alerting me that it cannot find cv2 (is that the virtual environment?) 

 

 

Anyways, my short-term goal is to create a Halloween project within the next few weeks - It will involve a 'floating eyeball' that 'follows' a person based on camera tracking. I found some open source python code that another person is using for a 'pi nerf turret' that I (think) I can use to actuate stepper motors to pivot the eyeball and camera. I may need to edit a little of the code, but the bulk of the work I believe is already done for me by somebody who knows what they're doing...LOL! I have no idea if I will be able to do this with my severely limited python and zero OpenCV knowledge, but it will at least be a fun learning experience. If I am somehow able to pull this off, I will definitely post pictures and video! Thanks!

[tikker]

1 hour ago, steelo said:

On another note, for some reason, I'm having to execute these .py demos in terminal - If I try running from within Thonny, I'll receive numerous errors on the bottom pane alerting me that it cannot find cv2 (is that the virtual environment?) 

Apparently it comes with its own Python, so it could be using that in preference over your system Python, for which you have installed OpenCV. You can probably change it in the settings somewhere. Terminal is much easier imo anyway

[steelo]

47 minutes ago, tikker said:

Apparently it comes with its own Python, so it could be using that in preference over your system Python, for which you have installed OpenCV. You can probably change it in the settings somewhere. Terminal is much easier imo anyway

Thanks. That's what I was thinking, now do I need to worry about the 'virtual environment' many of the other tutorials stress you work in or is this its own virtual environment? I assume OpenCV and all of its dependencies are installed correctly, as all of the demo projects seem to work fine in terminal?

 

Also, one of the steps in that tutorial linked above calls to make an 'empty' file and create a 'shebang' (I think that's what it's called) - #!/usr/bin/python. I assume so you can make the .py files executable? I followed this step and then deleted it and the projects still run, so I'm not sure of the purpose of this...

[tikker]

1 hour ago, steelo said:

Thanks. That's what I was thinking, now do I need to worry about the 'virtual environment' many of the other tutorials stress you work in or is this its own virtual environment? I assume OpenCV and all of its dependencies are installed correctly, as all of the demo projects seem to work fine in terminal?

The virtualenv is yet another thing. What I mean is Thonny has its own Python included, from its description, i.e. it is not using the same python as when you type "python" in your terminal.

 

1 hour ago, steelo said:

Also, one of the steps in that tutorial linked above calls to make an 'empty' file and create a 'shebang' (I think that's what it's called) - #!/usr/bin/python. I assume so you can make the .py files executable? I followed this step and then deleted it and the projects still run, so I'm not sure of the purpose of this...

You can always execute your script as "python myscript.py". What the shebang line does, is enable you to execute your script as "myscript.py" (if you can't, then a "chmod u+x myscript.py" in the terminal will make it executable).

[steelo]

57 minutes ago, tikker said:

The virtualenv is yet another thing. What I mean is Thonny has its own Python included, from its description, i.e. it is not using the same python as when you type "python" in your terminal.

 

You can always execute your script as "python myscript.py". What the shebang line does, is enable you to execute your script as "myscript.py" (if you can't, then a "chmod u+x myscript.py" in the terminal will make it executable).

Ah, thanks again...so, if I'm understanding you correctly, what the shebang line does is eliminate the need to have to type 'python' before 'myscript.py'-  It kind of treats it like an executable file.

 

 Tonight I'm going to play around with controlling servos. I purchased a servo controller HAT but from what I've read, that may be unnecessary. I've found a few tutorials where they directly connect a servo to an i/o pin for the signal and an external 5v power supply (which may or may not be a very bad idea)

 

Anyways, if I can gain an understanding of how image processing and i/o interaction works I just may be able to piece them together in simplistic form 

 

[tikker]

4 minutes ago, steelo said:

Ah, thanks again. Tonight I'm going to play around with controlling servos. I purchased a servo controller HAT but from what I've read, that may be unnecessary. If I can gain an understanding of how image processing and i/o interaction works I just may be able to piece them together 

 

A HAT is definitely recommended for servo stuff. The Pi just isn't that good at that.

[steelo]

3 minutes ago, tikker said:

A HAT is definitely recommended for servo stuff. The Pi just isn't that good at that.

Thanks again for your valuable input! I've read that it can be kind of flaky directly connecting servos. I know you can't just feed it a signal, it has to be a pulsating signal (50khz, if I remember correctly)

 

I'm glad then that I ordered a HAT

 

[tikker]

1 minute ago, steelo said:

I've read that it can be kind of flaky directly connecting servos. I know you can't just feed it a signal, it has to be a pulsating signal (50khz, if I remember correctly)

 

I'm glad then that I ordered a HAT

 

The reason is that the Pi has other things to worry about as well running your program, doing OS stuff etc. whereas that HAT is built for one thing: sending PWM signals to the servos. You can do it from the GPIO pins on the Pi directly (I've done it), but it wasn't very stable (lots of jitters) and when under heavy load it started missing steps/ticks or slowed down.

 

You use the width of a pulse to tell the servo which direction to go in. 50 Hz means you send a pulse 50 times per second. You then change how long that pulse is on in milliseconds (the width) to tell where it needs to go.

[steelo]

13 minutes ago, tikker said:

The reason is that the Pi has other things to worry about as well running your program, doing OS stuff etc. whereas that HAT is built for one thing: sending PWM signals to the servos. You can do it from the GPIO pins on the Pi directly (I've done it), but it wasn't very stable (lots of jitters) and when under heavy load it started missing steps/ticks or slowed down.

 

You use the width of a pulse to tell the servo which direction to go in. 50 Hz means you send a pulse 50 times per second. You then change how long that pulse is on in milliseconds (the width) to tell where it needs to go.

I'm sure since motion tracking already maxes out the pi, it would be a nightmare controlling 2 servos without it.

 

These are the guides I plan on using (I've noticed in the first tutorial he's not using a HAT so I don't know if the code would have to change) I know I don't only want to track an object of a certain color so I think I would need to remove the 'mask' he uses in his code. The second link shows that he set the camera to grayscale to track an object and attempt to center the camera on it.

 

https://www.hackster.io/mjrobot/automatic-vision-object-tracking-5575c4

https://www.hackster.io/hackershack/raspberry-pi-motion-tracking-gun-turret-77fb0b

[steelo]

I know the set up was bad, but I just wanted to see this work. I was using a rpi zero, an old webcam and a spare R/C car servo. Tracking was very slow and jerky and I was having issues with VNC viewer disconnecting because the rpi was so bogged down. I am still waiting on a stepper motor, rpi3 and a HAT which (hopefully) should make the tracking run more smoothly and not look so jerky. Also, I wouldn't use this particular code for my project as it masks all but objects of a certain color. There appears to be other code out there that will grayscale the video and sharpen contours to track better.