Raspberry Pi Camera

Bernard

Well-known member
Staff member
HI.
2 or 3 years ago i work on raspberry Pi and camera vision management.
OS was Buster 32 bits and everything work very well:
On a Pi3B+ or Pi4
Streaming video use 2 or 3% of CPU and object detection useless than 70% of CPU at 1.4 fps on Pi3B+

But Now, new test on the last Pi OS bookWorm 64 bits (also on bulleyes) Picamera change to Picamera2 (new version) and i can't have a correct result.
Streaming video use 12 or 13% of CPU :mad: and i fail to install tensorflow

So help from people with more knowledge on Pi are welcome;)
 
Hi, thanks for your information.
so if I compare your results with my tests, near the same. Yes since bullsey picamera is not supported and picamera2 has to be used. But there are a lot of changes form picamera to picamera2. May be that is the reason why tenserflow will not work. I never worked withe tenserflow before.
What I can do is to setup a tenserflow enviroment under buster, with your support and than i will try to migrate this to bookworm, but I can do this not before middle of February. I found also a lots of posts in the rpi forum from people they runs in Problems with tenserflow and bookworm, but also some youtube how to install, but I didnt dive deep into that.
 
What I can do is to setup a tenserflow enviroment under buster,
I describe all the setup on the wiki, but you need buster to have the picamera support and not bulleyes.
code
Maybe you need to follow all the setup and into config.py set DueConnectedOnPi to False to test without mower.
It's take time to setup everything, but you can test the result with only a Pi3B+ and a picamera on table.
Start piardu and into video menu click on start vision ,after 1 minutes you need to see the live video and picture are recorded on each detection.

Unfortunatly i can't find a way to directly share the SD Card image (32GB) file export from win323diskimager.
 
Thanks if this is okay for you I will start with the test end of next week. I have 2 pi4 and can put buster to them and I think I will get it up and running. Then the next step follows, we also can move this discussion in the mower forum, that all the member ca participate.
 
Here ist a Image with pre Installation
Yes i use this image, but the result is actually not what i expected.
Work on PI4 , but always slower than the buster one with Picamera.
And i stop to work on bookworm, Not yet stable enought.
 
May be this is the reason why bookworm (if you use the official Pi Imager) is only provided in the selection box for Pi4 and Pi5.
some Days ago a made also the experience, that RDP is not running propper on bullseye and bookworm. Hope we don't run in a "microsoft
desaster with every new update it don't runs faster"
 
Back
Top