I think I might have some user error or misunderstanding on my part. When I use your provided test, It reads it just fine both cpu and gpu. I had blue iris set to zoom on object. Removing that seems to have fixed my issue. My neighbor has a visitor that is parking right in my shot since yesterday so it's not an ideal situation for me. But I did just get a plate read and in the BI log I now see the plate. But if I take this image and run it through the Codeai explorer it doesn't read it at all.
Blue Iris and CodeProject.AI ALPR
- Thread starter MikeLud1
- Start date
Hey @MikeLud1 I see this once in a while (maybe 1 in 1000 plates) where spaces in the middle causes the OCR to reverse the order, it only happens when there is a space in the middle. Any idea what's causing it? I'm running the current version on an RKNN. It doesn't happen every time either, these reads are within a min of each other on different cameras (the first and second are on the same camera, vehicle just moving in different direction).
Last edited:
This happens when the OCR detects the characters as two separate detections because of the space. I have not look at the code recently, I will see if I can fix it.
I was going to say maybe the AI suffers from dyslexia sometimes but now based on Mike's response I have to respond that the OCR may be getting old. It has to reread things over and then mixes it up just like people. LOL
Best ALPR ever LOL.
This doesn't show up in the list of plates, but somehow got a great image of someone driving at night LOL
This doesn't show up in the list of plates, but somehow got a great image of someone driving at night LOL
I recently installed a Coral.AI and got the drivers working in my server. It works really well but I am getting the following error that someone else got. I set the model to be medium and that does NOT work for me. Any other setttings I should try?
github.com
Small modules ALPR stops working · Issue #267 · codeproject/CodeProject.AI-Server
Server Info CodeProject.AI 2.9.7 Docker 4GB Intel integrated GPU) Describe the bug When using Coral AI or other Yolo models with SMALL, license plate doesn't work: 12:06:30:License Plate Reader: [K...
github.com
HI! looking for some advice. I have everything working well EXCEPT.. CPAI is reporting plate numbers in the log but they are not appearing on the alerts in BI. I know I am doing this to myself somehow, just cant figure out what I did..
iwanttosee
Pulling my weight
I think there is a memory leak problem with the python code.
I have an automation to restart my Debian docker if the memory gets to 95% full.
I have an automation to restart my Debian docker if the memory gets to 95% full.
I too have experienced what appears to be a memory leak in CodeProject.AI, but I don't think we (or at least I) can know for certain whether it's a code problem in CodeProject.AI itself or if BlueIris (or any other client) is not closing a connection that it should.
I am not trying to hijack this thread, but I will reference a different thread that I started with some of my environment details & observations, rather than copy/paste/duplicating it here: CodeProject.AI 2.9.5 w/YOLOv5.NET (GPU/DirectML) memory growth/leak
I am not trying to hijack this thread, but I will reference a different thread that I started with some of my environment details & observations, rather than copy/paste/duplicating it here: CodeProject.AI 2.9.5 w/YOLOv5.NET (GPU/DirectML) memory growth/leak
iwanttosee
Pulling my weight
Thanks. Please let me know if you need any information from me.
My CodeProject.AI runs inside Docker within a Debian 12 VM hosted by Proxmox.
I was having perhaps some high CPU with Codeproject.AI after a few days. Last week I set a crontab to restart the container at 3 AM. I didn't watch the memory usage.
Been thinking about building a new machine to run everything rather than the 3-4 machines I have now. I have Codeproject.ai working in Docker and pass through the coral.ai but that doesn't help with ALPR. I have been thinking of getting a NVIDA card to accelerate ALPR and object detection. I don't want to drop $1000 on that. What card are you guys using? I also want to do CUDA acceleration for TTS.
I'll probably get a 24 core i9 with Quicksync for the plex container....Want to get one that does ECC.
I'll probably get a 24 core i9 with Quicksync for the plex container....Want to get one that does ECC.
iwanttosee
Pulling my weight
A micro OptiPlex 7070 running i9-9900T with 32GB of RAM running Proxmox hosting 4 VMs: Windows 10 BI, Debian Linux, Home Assistant, and another Debian running Docker hosting CodeProject.AI & License Plate Dashboard.
I use a Synology NAS for storage.
4 LPR cameras, 8 exterior cameras, and 3 interior cameras. No add-on video card and 6 CPUs assigned to CodeProject.AI for object and license plate detection.
The computer, synology and a few routers use less than 100 watts according to my APC ups.
Last edited:
I am using Codeproject.ai on an old Xeon E3 and my CPU times are about equivalent to yours. About 500-600 ms for a plate.
I use my ALPR data to open the garage doors, so latency is somewhat of an issue. I think I analyze like 30 frames everytime a car drives by. I am hoping to get that down. I've seen people get into the 40ms range. 100ms would be perfect for me.
I am currently using ESXi and running docker on a Ubuntu instance. I used to virtualize windows on a VM but I added too many cameras (before substreams). I've considered Proxmox for a new build, but I am leaning towards a baremetal Linux and just Virtualizing windows within that environment. If I have a NVIDIA card and a Coral.AI having fewer layers of abstraction makes things easier. I use substreams so Blue Iris would just be chugging CPU cycles and that would be my only VM.
I use my ALPR data to open the garage doors, so latency is somewhat of an issue. I think I analyze like 30 frames everytime a car drives by. I am hoping to get that down. I've seen people get into the 40ms range. 100ms would be perfect for me.
I am currently using ESXi and running docker on a Ubuntu instance. I used to virtualize windows on a VM but I added too many cameras (before substreams). I've considered Proxmox for a new build, but I am leaning towards a baremetal Linux and just Virtualizing windows within that environment. If I have a NVIDIA card and a Coral.AI having fewer layers of abstraction makes things easier. I use substreams so Blue Iris would just be chugging CPU cycles and that would be my only VM.