[tool] [tutorial] Free AI Person Detection for Blue Iris

I have a lot of my cameras with images of stationary (permanent) cars in. Will the AI learn that these aren't alerts?

Well yes I've been thinking about this issue. The Deepstack AI won't learn because it is not possible to actually train the software, that's a shame. As DeepquestAI can only tell the AI Tool "Hey yeah I see 3 cars, an elephant and a person" it's difficult for me to resolve this problem myself, without access to the internals of DQAI.

Maybe I will implement something like this: Every x minutes, a still without relevant content is taken and analyzed. It is saved which and how many objects were found in this reference image and where they were located. The following alert images then are checked and only if there are more or less objects than in the reference image or if they are at different places, an alert is triggered.

This is something that MIGHT be added in the distant future.

Regarding the console size: yeah I know it can't be resized, but that's purposely because I don't know how to design a UI resizable and it would look terrible. But I thought it's a background task, so who cares.
 
Last edited:
  • Like
Reactions: MnM
I'm getting false positives but I'm getting there. Just like to sort how it will run automatically on computer startup. I also want to be able to detect cars when there are cars parked in the field of view. Would be nice to be able to minimize the server screen. Still learning.

Clicking the minimize button will minimize the program to the tray, double click the tray icon and it's visible again
 
How do you get deepstack to automatically start when you start the computer?

It is not possible to autostart DeepQuestAI at the moment. I anticipate that the great developers at DQAI will change this in the future :). Meanwhile, if you need autostart you might want to use the Docker version of DQAI, detailed setup instructions can be found in the spoiler of the deprecated versions in the first post. In my experience this comes at the price of slower analyses.
 
It is not possible to autostart DeepQuestAI at the moment. I anticipate that the great developers at DQAI will change this in the future :). Meanwhile, if you need autostart you might want to use the Docker version of DQAI, detailed setup instructions can be found in the spoiler of the deprecated versions in the first post. In my experience this comes at the price of slower analyses.

I think the models on Deep Quest still need some work. For a camera I have mounted at the corner of my house. You can clearly see humans but in a few cases they aren't detected. I'm not fully confident in this yet but as it is now, it's still working great and saving me from a tonne of crappy false alerts.
 
I think the models on Deep Quest still need some work. For a camera I have mounted at the corner of my house. You can clearly see humans but in a few cases they aren't detected. I'm not fully confident in this yet but as it is now, it's still working great and saving me from a tonne of crappy false alerts.

To be quite honest, I have great respect for the people who developed this software. I haven't done much background research, but it seems to be a small startup from Nigeria, quite amazing.
In the end this does not want to to be an all-curing panacea, but it will save us from tons of false alerts. ;)
 
  • Like
Reactions: MnM and Forid200
Would it help you if certain areas oft the images were ignored by AI Tool?

Only if the horse stays still. :lmao: It could help though with the parked car issue though.

Its a shame it not self learning- i thought thats what AI was.

Great program- I'm loving it. I had that many alerts that I didn't send them to my phone but now it's possible.:)
 
  • Like
Reactions: GentlePumpkin
To be quite honest, I have great respect for the people who developed this software. I haven't done much background research, but it seems to be a small startup from Nigeria, quite amazing.
In the end this does not want to to be an all-curing panacea, but it will save us from tons of false alerts. ;)

Hmm this is definitely something I'll follow. An AI company from Nigeria is nearly unheard of. I hope it continues to improve as time goes on. This AI proves that you don't need a tonne of hardware, and that AI can function locally.
 
Only if the horse stays still. :lmao: It could help though with the parked car issue though.

Its a shame it not self learning- i thought thats what AI was.

Great program- I'm loving it. I had that many alerts that I didn't send them to my phone but now it's possible.:)

I wonder what the Blue iris offering will be...
 
  • Like
Reactions: DwightMorgan
Would it help you if certain areas oft the images were ignored by AI Tool?

Yes it would!! I think it might imrpove response time? (Less area to look at).
In certain situations it will help. I get alerts (of cars for example in the street for example) and it just so happen that there are pedestrians across the street walking, at the same time.
The alert is triggered by the shadow of the car that goes on my driveway when the car goes past, so everything in the generated picture is irrelevant. But because, by chance there are pedestrians in the picture (far away from the zone that I am really interested in) I get an alert.

So yes the capability to have same/similar ignore zones as per BI settings of the original camera will help to reduce the alerts even further.

Thank you again!! Love the software. Installed it on other BI servers that some of my family members have for their properties. They love it too.

My stats so far are:

Alerts: 200
Irrelevant Alerts: 1208
False Alerts: 172

:)
 
update v1.53
  • privacy mask feature
  • clean redesign of the history tab
  • detected objects now can be marked with a rectangle around them (replacing the image cutout feature)
  • AI Tool now can be resized/maximized
  • some more visual enhancements
 
  • Like
Reactions: Tinbum and Forid200
update v1.53
  • privacy mask feature
  • clean redesign of the history tab
  • detected objects now can be marked with a rectangle around them (replacing the image cutout feature)
  • AI Tool now can be resized/maximized
  • some more visual enhancements
Great work!! However strangely enough randomly my Deepstack server stopped working today :/

Tried an Ubuntu docker install too and that bought along it's own issues such as the threads aborting.

EDIT: Reinstalled Deepstack AI server, rebooted all working again. Strange one, it completely stopped responding to the AI tool.


Thought this was a little funny - thought the telephone pole was a person xD
QRdVGdv.png
 
Last edited:
Thank you. Appreciate all your hard work. As mentioned before this is just getting better and better. :)
 
I'm trying to implement the older command line Docker version but I'm having trouble figuring out the right syntax for the config.txt file. I have three cams and three separate alerts, but when I list the 3 trigger URL's in the config file all three alerts fire. How do I get the config set so the 3 cams have 3 independent alerts?
 
Sorry I don't use the docker version.

Is it possible to reset the stats? I see they are stored in the individual camera file but they seem to be admin locked.
 
Last edited by a moderator:
Bit of a strange one here. see attached. The vehicle is masked out (150 op) but AI still detects it. ??

edit Possibly because it's right in the corner /edge of the image?? I have checked Images are the same resolution.
 

Attachments

  • drive.jpg
    drive.jpg
    293.4 KB · Views: 209
Last edited:
Tried shutting it down but still couldn't alter. so went into properties/ security/ edit and changed permissions and that did it.