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

Read the comments in this thread:


The cameras AI will work with Blue Iris. What you cannot do prior to DeepStack was do a search by humans or faces with BI. That is the type of stuff you could do with the NVR AI, but you can trigger for those events in BI.

But if all you care about is triggering the camera to send you an alert or push or SMS for human and/or car and do not need all the other bells and whistles that DeepStack offers, this works very well.

How to use the AI in the camera instead of BI motion detection:

Go to Smart Plan and select IVS and hit save.

Then go to IVS and add an IVS rule. Use Intrusion and select the appears and crosses boxes. Draw a box/outline around the area you want it to trigger for, but try not to do the full frame to give the camera time to recognize the object. Then check the box for human and/or vehicle. There is no sensitivity settings in this setup.

Leave min size to 0,0 and do not do any other changes and hit save.

Make sure both motion detection and Smart Motion Detection are not checked in the camera.

Then in BI you need to set it up in two places. In the motion tab, select the camera's digital input or motion alarm.

Uncheck the BI Motion Sensor.

Then go back to the setup screen where you key in the camera IP address and user and password and down in the lower left (or up a little depending on the version you are running), will be the checkbox for pull ONVIF triggers.

I ran a clone camera for awhile using BI motion to compare it to the ONVIF triggers to confirm I wasn't missing anything. Now for those cams I just use ONVIF triggers.

Having the camera do motion detection instead of BI does free up CPU as motion detection is one of the larger CPU drains.

I repeat, do not have Motion Detection and Smart Motion Detection turned on in the camera.

This will eliminate your false triggers.
 
I just upgraded to the latest AI Tool version.
Can anyone please explain what a "Meat Popsicle" is?
Also, when I add a new object, how do I train/set deepstack to know what I mean?
Thanks.
 
Does anyone know the correct syntax for running the Deepstack dark model (dark.pt) in a Docker / Windows environment ?

Examples are shown for both Windows and Linux / Docker but I need a Windows / Docker example.
 
Does anyone know the correct syntax for running the Deepstack dark model (dark.pt) in a Docker / Windows environment ?

Examples are shown for both Windows and Linux / Docker but I need a Windows / Docker example.

If you find the syntax please post. I would also like to know how you end up using the dark mode in AITool (active time range, linked server, refinement server).
 
If you find the syntax please post. I would also like to know how you end up using the dark mode in AITool (active time range, linked server, refinement server).

I have reconfigured my system to use the Windows version of DeepStack and have now tested the dark.pt night-time model. I had great expectations especially since the latest version of Vorland-cd AITool has the ability to switch models on a dusk-dawn basis.
Alas my experience has been disappointing to say the least. The dark version recognised me about 5% of the time versus 90% using the standard release which makes it unusable for me. Needless to say your mileage may vary. The dark version frequently recognised me as a table lol

FYI I have attached the image that I tested the dark.pt model with. Hmmm might be an alien but definitely not a table :lol:
 

Attachments

  • Dark Test Image.png
    Dark Test Image.png
    5.5 MB · Views: 71
Last edited:
Is anyone else getting this DeepStack error in AI Tool when using more than one instance of DS? It seems like the file its looking for gets replaced by another file by one of the other instances. All instances still work after throwing the error. I'm just trying to see if I can stop it from happening.

Code:
Process Process-1:
Traceback (most recent call last):
  File "D:\obj\windows-release\37amd64_Release\msi_python\zip_amd64\process.py", line 297, in _bootstrap
  File "D:\obj\windows-release\37amd64_Release\msi_python\zip_amd64\process.py", line 99, in run
  File "C:\DeepStack\intelligencelayer\shared\detection.py", line 138, in objectdetection
    os.remove(img_path)
FileNotFoundError: [WinError 2] The system cannot find the file specified: 'C:\\Users\\admin\\AppData\\Local\\Temp\\DeepStack\\571385ee-582a-43e3
 
Is anyone else getting this DeepStack error in AI Tool when using more than one instance of DS? It seems like the file its looking for gets replaced by another file by one of the other instances. All instances still work after throwing the error. I'm just trying to see if I can stop it from happening.

Code:
Process Process-1:
Traceback (most recent call last):
  File "D:\obj\windows-release\37amd64_Release\msi_python\zip_amd64\process.py", line 297, in _bootstrap
  File "D:\obj\windows-release\37amd64_Release\msi_python\zip_amd64\process.py", line 99, in run
  File "C:\DeepStack\intelligencelayer\shared\detection.py", line 138, in objectdetection
    os.remove(img_path)
FileNotFoundError: [WinError 2] The system cannot find the file specified: 'C:\\Users\\admin\\AppData\\Local\\Temp\\DeepStack\\571385ee-582a-43e3
I had a similar issue when I ran DeepStack under Windows. I resolved it by reverting back to running DeepStack in a Docker environment. Your mileage may vary!
 
Is there a reliable link to the currnet compiled version of the VolronCD releases? I use GitHub quite often, but I can't find where the releases are being stored on this one :(
 
I have reason to believe the project is dying. I hope I am mistaken.

Given that there currently is no known reliable/working deepstack windows GPU version that I am aware of I decided to get the code and do it myself.

What I found was a surprise to me and let me just say that things need to be cleaned up real fast in order to have any life expectancy.

I have reached out to @john on github regarding the state the current code is in but havn’t heard back yet.

If there is anyone with a working windows GPU version (preferably a recent one) please zip it up and share it. I can decompile the code if need be and start from there in an effort of cleaning house.
 
Last edited:
  • Like
Reactions: clowg and WimNL
i have a problem with AI Tool.. im working with clone aicams. Blueiris send me a video with mail... but it is over 20-30 sec to late ...
same with alert image. Recording delay 8 sec.
 
That could be an issue with your mail more than AI tool. Check the timestamp in the BI log when it sent the mail versus when you received it.
 
just a heads up. My issues were never with the code from vorloncd. My issues are with the deepstack code. Through help from someone one another forum I have gotten things to at least run on the windows GPU 2021-2-1 version. I will try and see if I can get compile the the 2021-8-1 version and if it runs. Should anyone have an interest in that then please let me know why so I can be motivated to push trhough
 
That could be an issue with your mail more than AI tool. Check the timestamp in the BI log when it sent the mail versus when you received it.
hmm its same with "alert video" in blueieris, blueiris app, and mail. I think aitool send to late and send the confirm time after deeptstack. I need alert time from Blueiris alert for deepstack.
But time is not in the tigger link "[BlueIrisURL]/admin?trigger&camera=[camera]&user=[Username]&pw=[Password]&flagalert=1&memo=[summary]&jpeg=[ImagePathEscaped]"
I have activate "buffer recording" but its not working in alarm videos.


Edit:
i have check. time on alert pic ( ai tool) 08:46:33 and in blueiris Alert: 08:46:38...
 
Last edited:
Is the deepstack gpu version still only working on nVidia cards? I have an AMD firepro and would like to try that sometime (only have 3 cameras).
 
Also I've been struggling to combat false alerts with really heavy fog. First time in 2 years (had my set up just right) and lately we've had a lot of fog at night time which looks like slow falling snow in front of my cameras.
Reduced the env variable to Medium from High. Also the break time down to 1.5 seconds.

cpu was working hard last night analysing images for 2-3 hours early in the morning. Still triggering.
 
Last edited:
Is there any way of setting up a custom model as a refinement server to only run if the main server does not find any relevant objects?
I see there is an option to refine but only if the object IS found in the first server. The opposite would be good imo.

1629749721954.png