[SOLVED] Raspberry Pi 3 for image processing

[Thom13]

Hi, I will be working on a project that will require image processing. My question is, Is Raspberry Pi 3 enough for image processing or should I get Raspberry Pi 4? And what will be the benefits of Raspberry pi 3 vs 4 when dealing with a situation like this? Thanks.

 

[kanewolf]

The PI 4 1GB is the same price. It has faster networking. There should be NO reason notto get a 4.

 

[OldSurferDude]

In 2011 I did image processing on an EEEPC running an archaic linux. This configuration is about the same as an RPi.

The things that affect processing speed are RAM, disk access, then CPU. Your processing speed will be exponentially slower the further the disk is away from the CPU: internal vs USB vs NAS vs Cloud.

Pi 4 will actually be a little more expensive because, I read, it absolutely needs heat sinks and probably a fan, where Pi3 works just fine if you have it in free air.

As always in computing, is it speed or cost? Choose one. Having it look "cool" will cost more.

 

[Thom13]

The PI 4 1GB is the same price. It has faster networking. There should be NO reason notto get a 4.

Hi, thanks for the reply. Will the 4GB Ram version of Raspberry Pi 4 make a huge difference compared with the 1GB model? Or will the 1GB be fine?

 

[kanewolf]

Hi, thanks for the reply. Will the 4GB Ram version of Raspberry Pi 4 make a huge difference compared with the 1GB model? Or will the 1GB be fine?

I have no idea. Obviously you get more buffer space. Will that significantly improve your perfoirmance? No idea. You would have to find benchmarks with the software you are using to answer that question.

 

[kanewolf]

Is your plan to create a cluster our of PIs and distribute your processing across the cluster ?

 

[USAFRet]

a project that will require image processing

What level of "image processing?
There is a large difference between editing a couple of jpgs from your phone, and running facial recognition for 500,000 security cam images, or analyzing xray images that are 20GB each.

 

[Thom13]

Is your plan to create a cluster our of PIs and distribute your processing across the cluster ?

By cluster, you mean distributing the work load across multiple RPis? (Sorry I am kinda new in these things.) If yes, then its not exactly what I am thinking of. What I plan to do is for all the workload to be handled by 1 RPi.

Sorry I am new with Raspberry Pi and SBCs. Basically my project is, I have a web server or web app in Raspberry Pi and a camera that will be constantly getting and comparing images from the web server. And if a certain image fits, then it will configure a servo motor. At least that's the idea I have in my mind. I am yet to perform my first experiment regarding this problem. First I have to buy a RPi but I am having problem choosing the right one.

 

[Thom13]

What level of "image processing?
There is a large difference between editing a couple of jpgs from your phone, and running facial recognition for 500,000 security cam images, or analyzing xray images that are 20GB each.

Hi, thanks for the reply. I plan on extracting sequence of letters and numbers on the image. Plate numbers to be exact.

 

[OldSurferDude]

Ah! If you're talking processing speed, and I think you are, 2GB is essential, 4GB will be even better. This reduces disk swapping (virtual disk) Also, make sure your SD card has the fastest R/W speed.

The cluster idea is a good one, again the cost (including development time) vs minimizing processing time.

But if speed is really important, start with a PC motherboard with a really fast processor, add a boatload of RAM and an SSD drive. If you're doing automated processing, don't you don't need a high powered graphics card, If you are doing the tweaking, then a good graphics card is essential.

I'd go with a minimal Ubuntu. My application was automated so I used Imagemagick which is a very powerful command line software tool.

 

[USAFRet]

Hi, thanks for the reply. I plan on extracting sequence of letters and numbers on the image. Plate numbers to be exact.

ANPR (Automatic Number Plate Recognition)
Multiple solutions already exist for this.

Are you rolling your own software, or using an available library?

What is the focus of the project?
To prove the concept of using R Pi for this, or is it to do the actual plate recognition?

 

[Thom13]

ANPR (Automatic Number Plate Recognition)
Multiple solutions already exist for this.

Are you rolling your own software, or using an available library?

What is the focus of the project?
To prove the concept of using R Pi for this, or is it to do the actual plate recognition?

I might use available solutions and libraries. This is project is for my class on embedded systems. Basically we want to somehow kick start ourselves on SBCs.

 

[USAFRet]

I might use available solutions and libraries. This is project is for my class on embedded systems. Basically we want to somehow kick start ourselves on SBCs.

OK, so the Pi is actually part of the project.

Look at what software you might want to use, and see what its requirements are.

 

[Thom13]

OK, so the Pi is actually part of the project.

Look at what software you might want to use, and see what its requirements are.

By software, do you mean the OS or the actual software on detecting plate numbers?

 

[USAFRet]

By software, do you mean the OS or the actual software on detecting plate numbers?

Yes...

Obviously, the Pi specific distro you choose will run on some level of Pi.
The plate recog software, however...that may have specific minimum requirements.

 

[Thom13]

Yes...

Obviously, the Pi specific distro you choose will run on some level of Pi.
The plate recog software, however...that may have specific minimum requirements.

Alright thanks. I'll do that. I might use available libraries and program the plate recognition software my self if I feel doing stuff though, but I will still consider the other factors. Thanks, I appreciate it