New User

[shipracer_camera]

Hey Guys, New user to blue IRIS and code project AI. I'm very techy.

For the most part everything is working, objects are being recorded. Im stuck on a couple things (feel free to ignore or point me to read some more, im fresh to this).

I cant get codeprojectAI

http://localhost:32168/vision.html

to read or guess a license plate on a clear as day image. It does see it as an object DayPlate, but its not able to read the plate numbers.

I've tried other images, installed nvidia cuda manually.

Also I keep getting Alert Cancelled Occupied (I have not been able to figure out).

Thanks for all the help and info so far.

 

[wittaj]

Post a screenshot of the field of view of the camera you are trying it for.

Maybe you are trying to do too much with the field of view.

 

[shipracer_camera]

I get what your saying, Its a reo-link dual camera 180 degree camera.

I know now from reading and videos that getting plates requires a bit of effort (possibly dedicated cameras pointed and focused/aimed perfectly, or cloning cameras and assigning license plate functions to only one).

Though I don't really need any license plates from this camera, there are just quite a few times a day plates are clear as day, easy to read by the human eye.. but the AI only looks initially on the movement and not the whole time or similar (I'm a complete newbie here).

but in my case, I cant even get a still image, with a clear license plate... to read in the openai vision tool directly .. It sees it as a day plate but I cant get it to ever read it.


Is there anyway to disable the alert canceled option completely, so that all alerts occupied or not come through (until I learn or tune things better)? Or is that the right route?

 

[Swampledge]

It’s not the right route. There is a difference between a trigger and an alert. A trigger means an event (such as movement) has occurred. An alert is Blue Iris alerting you that the event has satisfied some type of user defined criteria, such as a certain amount of movement, or movement within or between user defined zones, or that the AI has assessed the possibility of an object being one of the objects you want to be alerted for. If you want an alert for every trigger, you don’t use AI.

Alert cancelled simply means that AI decided not to alert you because the images associated with the trigger don’t contain the object(s) you were looking for.

 

[wittaj]

Yeah a 180 camera will not read plates unless the car is within a few feet. Too far away and too wide an angle.

Ideally the field of view to capture plates shouldn't be much bigger than a vehicle.

 

[shipracer_camera]

Makes sense, but Im using a super clear image of the rear of the car you can read withtout any zoom, directly in openai vision.html manual tool and its not reading the plate. Shouldn't openAI be able to read it then? it sees the object as a day plate.

Is there a way to make the AI more accurate vs the length of time to process? Just more real images? IE: there was a lady walking by and it wasnt sure enough it was a person, so it didnt alert, but it clearly was a person the entire time she was in frame, but it only checks for a small duration and then its under the % of confidence to show.

or is there an aritcle or video anyone recommends to kick me in the head a little. Thanks for the replies.

 

[wittaj]

Just because it can tell it is a plate doesn't mean it can read it. The AI sees the back or front of the car and is looking for a square.

How close is the car to the camera?

Post a screenshot and we will be able to tell you if AI will read it.

But even if it can read it, a reolink can't run a fast enough shutter at night to capture a plate.

 

[shipracer_camera]

Yea, Im not really wanting license plate capture on this camera, I will buy a different/dedicated LPR camera (after reading more and asking more what you all recommend). I just figured OPENAI could read the plate if I could read the plate. Its a high res image focus'd on the back of the car. I see blurry as hell image reads, this is a crystal clear picture.

just trying to make sure I have everything setup/working right, I tried the different YOLO object detections, installed nvidia cuda manually.

thanks

 

[wittaj]

There is no way that is a reolink 180 camera picture.

 

[shipracer_camera]

No, that is a random picture from a photographer, Im just testing with a sample image, I used a snapshot from reolink (its the same thing, no plate number read).

but when I use the same image on plate recgonizer .. it reads the plate and a ton of other details... I feel like my settings or something is off, it should be able to read that image

 

[shipracer_camera]

Heres a reolink one just for reference (this isnt the cleanest one) but you can still read the plate with a human eye.

 

[wittaj]

Yeah it will have trouble with that. You need 60-80 ppf to reliably read plates and that will be nowhere close.

I can't read that plate without zoom.

You may get some during the day, but the percentage wouldn't be very high. And at night forget it.

The plate needs to be a lot larger in the field of view.

 

[shipracer_camera]

Yep. All of that makes sense and understood, thanks.

But why can it not just read the plate on a clear imagine manually. I still must have something wrong. Without blue iris, without reolink.

 

[wittaj]

Post screenshots of your CodeProject settings.

@MikeLud1 is the creator of the LPR for it so he will be able to see your settings and no immediately what the issue is.

 

[shipracer_camera]

Ohh great sure.

My issue and concern is not currently with blueiris and my code project settings, its with codeproject itself. My issue exsits outside of blueiris. With the manual image uploads and analysis's from

http://localhost:32168/vision.html

System info:
Server version: 2.2.4-Beta
Operating System: Windows (Microsoft Windows 10.0.19045)
CPUs: AMD Ryzen 5 1600 Six-Core Processor (AMD)
1 CPU x 6 cores. 12 logical processors (x64)
GPU: NVIDIA GeForce GTX 1660 Ti (6 GiB) (NVIDIA)
Driver: 545.84 CUDA: 12.3 (max supported: 12.3) Compute: 7.5
System RAM: 16 GiB
Target: Windows
BuildConfig: Release
Execution Env: Native
Runtime Env: Production
.NET framework: .NET 7.0.10
Video adapter info:
NVIDIA GeForce GTX 1660 Ti:
Driver Version 31.0.15.4584
Video Processor NVIDIA GeForce GTX 1660 Ti
System GPU info:
GPU 3D Usage 1%
GPU RAM Usage 3.1 GiB
Global Environment variables:
CPAI_APPROOTPATH = <root>
CPAI_PORT = 32168

 

[shipracer_camera]

Has to do something with CUDA, I finally found that you can switch to CPU vs GPU on the LPR module and it sees plates instantly.

I tried multiple 11.x versions, latest version. I must be doing something wrong.