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

This software is really EPIC !!!! thank you very much!!

I have a problem, I continue to receive the below message in the log. I cannot understand why? Someone had it and solved ?

[25.10.2020, 17:33:25]: ERROR: Can't write to cameras/history.csv!

Anyway, it is working fine except that Docker Desktop for windows continue to crash after 30 minutes of so.. already tried to re-install Docker and the components. I may need to try use docker on another VM.

1603663927811.png

1603664021114.png
 
i just installed the Windows version 1.67. Can someone give a high level explanation of what is generating the input files that AITool is working on? The motion detection is turned off on the BlueIris application, so I'm not sure what triggers the file creation.
 
No, you should have one. Put one back in there.
Well...that was easy. :)

Do you happen to have as easy a fix that'll put the image that triggered an alert into the alert sms?

(Like, if a "person" is detected, I really want to see that person when I get a text, not my empty front yard.)

Pleeeeease???
 
i just installed the Windows version 1.67. Can someone give a high level explanation of what is generating the input files that AITool is working on? The motion detection is turned off on the BlueIris application, so I'm not sure what triggers the file creation.

I'm still trying to figure it out too, but I think it's this:

Finally, go to Alerts and uncheck all motion zones. Makes sure that external is checked.

This is how I've set mine up; still trying to determine if this is correct according to the above.

1603666703723.png
1603666769968.png
1603666728398.png
 

Attachments

  • 1603666625453.png
    1603666625453.png
    11.3 KB · Views: 17
Yeah; this is totally confusing. :-(

I just walked out in front; no alert, nothing in the aitool history.
Back in camera settings, I checked the motion detection box and then walked back out front. I got 4 texts within about 5 seconds. none of them have me in the picture. ?????
I do have a 42 second clip, I show up 20 seconds into it (I expected that because of the settings).
As I'm typing this, I just got a 25 second clip a few seconds ago of a car driving by. But no alert...which seems correct since I'm not telling aitools to look for cars.
So now I'm getting recordings of all motion that's detected, and only getting alerted to motion that has people/cats/dogs in it. But the alerts don't contain a picture of what created the alerts.

My goals are:
To ONLY record when I get an alert that i've defined in aitools
To get a text that has a picture of whatever had caused aitools to fire the trigger

Why is this so hard?? What am I missing??? I've been an IT professional for literally 30 years, but this is apparently beyond me??

Ugh...I'm so frustrated with this...but there aren't really any alternatives, so I want to push thru and make this work as i expect it to.

i can't imagine trying to setup 9 cameras!!
 
Fantastic tool. Great work.

Any plans to have an option to replace the BI alert image with one edited by AI-tool to overlay detected objects onto the original image.

If not may I propose that as a suggestion for a future version.
 
Last edited:
  • Like
Reactions: Nierka
Yeah; this is totally confusing. :-(

I just walked out in front; no alert, nothing in the aitool history.
Back in camera settings, I checked the motion detection box and then walked back out front. I got 4 texts within about 5 seconds. none of them have me in the picture. ?????
I do have a 42 second clip, I show up 20 seconds into it (I expected that because of the settings).
As I'm typing this, I just got a 25 second clip a few seconds ago of a car driving by. But no alert...which seems correct since I'm not telling aitools to look for cars.
So now I'm getting recordings of all motion that's detected, and only getting alerted to motion that has people/cats/dogs in it. But the alerts don't contain a picture of what created the alerts.

I would try to focus your testing more. if it's just you, maybe walk out -> leave an item -> then see if your blank alert image is before or after your motion trigger event that you would like the image of. You can improve this further and get it down to how many seconds it is off. if it's 20, its probably related to your pre-trigger buffer setting. if its 2s, it may be the delay in response time that it takes deepstack to process (or delay it takes camera to trigger?) odds are it is a setting somewhere. but your deepstack times are pretty slow to be fair - remember, it can't make an alert until after it has fully processed the 1st (sometimes possibly 2nd) image and ID'd a person.
 
Here's an example timeline:

Aitool history shows 4 "Person" detections at 4:32:29, 4:32:32, 4:32:35, 4:32:38. All images are of yours truly in different "poses".
There's a recorded clip in BI from 4:32:02 to 4:32:44:
  • My front door (screen door) begins to open at 4:32:22.
  • I'm fully in the frame at 4:32:24.
  • I'm out of the picture and the screen door is closed (also now out of the picture) at 4:32:34.
I get 5 texts at 4:32:xx (my message app doesn't show seconds). All 4 texts have the same image, with the same time stamp: 4:32:22.

Next alert happened just a few seconds after the one above.
Aitool history shows 4 irrelevant Car images at 4:35:17, 4:35:20, 4:35:22, 4:35:25.
There's a recorded clip in BI from 4:34:56 to 4:35:21 where a car leaves the neighbor's yard and drives past our house from 4:35:14 to 4:35:21.
I get a single text at 4:35:xx (my message app doesn't show seconds). The time stamp on the image is 4:35:14.

Too tired right now to decipher what this all means; will get back on it tomorrow after I get home from work. :-)
 
I get 5 texts at 4:32:xx (my message app doesn't show seconds). All 4 texts have the same image, with the same time stamp: 4:32:22.

It sounds like it is showing the alert image as it should, just as the motion occurs from the screen door opening (or a split second before).

To make things easy, I'd also make sure that the time stamp you are using as a reference is placed by the camera (not BI or AI), and has been recently synced to your PC time in the camera web config. that way you know a certain time, is a certain time, and there's no way you can get some other image for a given timestamp. IIf you had this setup in your first example, this alert would have to show the screen door. If it doesn't, it is probably a system issue rather than a setting.

My point is that it seems to be triggering at the right times, so you should work to find out which direction it needs to move. Doing this will also make it apparent if the alert image occurs as it should (and just slightly off in timing) or if it is a problem with the alert image.
 
First have a look in the AITool input folders at the images you get in there. Are you in them? If not it's BI settings that want looking at. Are there 5 that are all the same?
 
This software is really EPIC !!!! thank you very much!!

I have a problem, I continue to receive the below message in the log. I cannot understand why? Someone had it and solved ?

[25.10.2020, 17:33:25]: ERROR: Can't write to cameras/history.csv!

Anyway, it is working fine except that Docker Desktop for windows continue to crash after 30 minutes of so.. already tried to re-install Docker and the components. I may need to try use docker on another VM.

View attachment 73371

View attachment 73372


I moved DeepQuest on a Photon VM on my esxi and it is now working perfecty. It also 600ms vs 3/4 second faster without almost resources.My blue iris is also on a Vm.
 
I'm still receiving all these notification

[26.10.2020, 15:07:08]: ERROR: Can't write to cameras/history.csv!
[26.10.2020, 15:07:08]: ERROR: Can't write to cameras/history.csv!
[26.10.2020, 15:12:08]: ERROR: Can't write to cameras/history.csv!
[26.10.2020, 15:12:08]: ERROR: Can't write to cameras/history.csv!
[26.10.2020, 15:12:08]: ERROR: Can't write to cameras/history.csv!
[26.10.2020, 15:12:08]: ERROR: Can't write to cameras/history.csv!
[26.10.2020, 15:12:08]: ERROR: Can't write to cameras/history.csv!

Someone solved this problem ?

As I read somewhere in the topic, it seem connected to when the pictures in the AiInput folder are deleted. Also if it doesn't make much sense as they are different folders.
 
Fantastic tool. Great work.

Any plans to have an option to replace the BI alert image with one edited by AI-tool to overlay detected objects onto the original image.

If not may I propose that as a suggestion for a future version.

Its already been done in the VorlonCD version.
 
My question is probably very basic and somewhat Mundane when compared to most on here.
Here is my issue. I use my Blue Iris cameras to catch alerts of animals and birds around my property as well as people and vehicles. And can set the sensitivity settings to get those things to trigger Blue Iris. But obviously this give numerous false alerts from rain, wind and spiders at nights.
So I tried Using VorlonCD with my Blue Iris setup (I use a Camera Clone in Blue Iris to supply the trigger image) and the latest Deepstack CPU Beta running in docker. (latest as of yesterday).
That all seems to work but I find VorlonCD shows False alerts except for when people are actually in the frame. I have Birds selected as well as People in VorlonCD
Squirrels/birds all trigger Blue Iris but after going through Deepstack/Vorlon I get False alerts and the Main Blue Iris camera is thus not triggered. Any ideas?
 
It's a different release
@Tinbum ref the newest release- is the link I quoted from @Jooced post here and above "the place" to find it or are there other locations? I asked B/C I see where people ref this Version vs that one, are they compiling their own? How do you know which version you are running? May be wrong but I have been opening the manifest file and looking for changes in there i.e. the one I am running right now is 1.71.0.34045 I had downloaded what I thought was a newer one but it said 1.67.8.33855 am looking in the right place?