Blue Iris and DeepStack ALPR

Thank you so much Mike for this amazing capability. Voting for the Exe as well :)
 
I just posted Version 3.0 in the first post.
 
  • Like
Reactions: looney2ns
@MikeLud1 this is excellent work, as always! Thanks for building and sharing!

A bit of a side question if you don't mind, I'm wondering if I would be able to display the plate characters on screen using BI's overlays, like I'm doing currently with the Plate Recognizer integration?

Example:

View attachment 114710 View attachment 114712
View attachment 114711
@aesterling I just posted Version 3.0 and this version can display the plate characters on screen using BI's overlays just like the Plate Recognizer integration
 
Try stopping the service (see Start Menu DeepStackALPR) after stopping the service try copying the ini file and repasting back in the folder as the same file name. After making changes restart the service (see Start Menu DeepStackALPR). Let me know if that works.

Thank you that worked. started the service back up now just have to wait for a car to come by
 
Thank you that worked. started the service back up now just have to wait for a car to come by
It looks like the software I am using to create the install exe (InstallForge) is saving all the file to the DeepStackALPR folders that requires admin rights to edit them. So if you need to edit the ini file open notepad first with admin rights then open the ini file. I will try to fix this tomorrow.
1643592541788.png
 
Nice work.

Is it possible to use other models with this addon?

I am using licence-plate model due to living in Denmark.

OR..

is it possible to use your models and tweak the “overlapping_threshold, min_conf_thres, min_len_thres, & plate_len_threshold” ? and which one to change(what do they do)?
 
Last edited:
  • Like
Reactions: looney2ns
Is it possible to use other models with this addon?
In V3.1 I will add the option to change the models used by the API
is it possible to use your models and tweak the “overlapping_threshold, min_conf_thres, min_len_thres, & plate_len_threshold” ? and which one to change(what do they do)?
Most of these settings will not help with the wide plates you have in Denmark. In Version 3.1 I am going to add a setting to help with wide plates
overlapping_threshold: Sometime DeepStack will find two or more characters in the same position. This setting is how many pixels from from each of the characters found before it will know it is the next characters. This will ignore the lower confidence for the overlapping characters.
min_conf_thres: Is confidence level for identifying the characters.
min_len_thres: Is the minimum number of characters that the license plate will have.
plate_len_threshold: Is the maximum number of characters that the license plate will have any more the API will in ignore the lower confidence for the extra characters. The license plate in Denmark looks to have 7 characters then decrease the number to the number to 7
1643636792239.png
 
What are the second numbers in the detection ex. [515]? Is it detection time?

I asume the first number are time between images from camera.

1643640156165.png
 
isn't it very slow?

I see other peoples time around 100ms.

There are so many factors coming into play - is it GPU or CPU version, how many models running on system, how many cameras trigger at the same time, etc. Anything less than 1 second is probably ok for most.
 
  • Like
Reactions: looney2ns
There are so many factors coming into play - is it GPU or CPU version, how many models running on system, how many cameras trigger at the same time, etc. Anything less than 1 second is probably ok for most.

I have CPU version with 1 model and I trigger overview and LPR camera at the same time. I use hardware decoding "intel VVP" on a i5 8gen
 
Is it possible to train my own DS model? I get a lot of cancelled alerts even with very clear plates in FOV.(at night)

Currently i am using licence-plate model but i was told that is was trained onlywith daytime images.
 
Well 3.0 seems to be working really good, have some trigger issues to figure out at night and a couple during the day large vehicles seem to get missed some times Fedex,ups trucks.
How would one set this up to do two cameras?