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

@Chris Dodge Will you be releasing an update soon? Every now and then AITools gets stuck on a .jpg and doesn't process any subsequent images. Thanks
 
Dear team,

I'm running BI cloned camera + volconCD.
Is there a way to run just single CAM and avoid having all false alert reported to BI

Thanks for your help
 
Dear team,

I'm running BI cloned camera + volconCD.
Is there a way to run just single CAM and avoid having all false alert reported to BI

Thanks for your help

Yes you can but you will need to setup BI to record continously and then set alerts so they can be cancelled. Presently I'm not personally aware of a way to achieve this without recording continously. Please chime in if anyone knows more.

I should add this is a moving target that many users are hoping to achieve. So tomorrow ..........
 
Last edited:
  • Like
Reactions: seth-feinberg
That is what 5.3.3 does. My substream is used for motion and now it grabs the image from the main stream and sends that to AITool folder.

under where I can find that option? to use substream as motion and take Main-stream's image
 
under where I can find that option? to use substream as motion and take Main-stream's image

It isn’t an option. When using 5.3.3 that is just want happens. But you need to ensure you have a pre trigger recording of a few seconds to ensure it can go back to get the image.
 
just changed over to VorlonCD version.. however no matter when a trigger happens, I never get it to send me a photo to telegram. I just get


Code:
TelegramText> Still in TELEGRAM cooldown. No image will be uploaded to Telegram.  (0.0641 of 0.0833333 minutes - See 'telegram_cooldown_minutes' in settings file)
[14.10.2020, 12:25:00.297]:              TelegramText> Still in TELEGRAM cooldown. No image will be uploaded to Telegram.  (0.0643 of 0.0833333 minutes - See 'telegram_cooldown_minutes' in settings file)

I know the telegram bot works because I had errors being sent previously because I had the wrong password in the camera trigger url and was getting those notifications no problem.

I have cooldown minutes on 0
 
Worked with Ken and found that the trigger URL to override the current image for the SMS is &trigger&jpeg=<PATH>
Found that using @Chris Dodge AI tool, and the original @GentlePumpkin, the trigger does not work on sending the image.
If you manually take out the URL that was attempted from the AI log, and put it in the browser it works just fine.
This also only works for the non-MergeImageAnnotations image file, so the original one captured and sent to AI. (not sure why the MergeImageAnnotations one wont work)
I am going to run some PCAPS if I cant find a way to get BI at a higher log level, to see if there is a difference on the receiving of the URL

Thoughts on a possible fix out the w.o log review.
If the merge step could be combined with the copy to file step and make a file in a directory then move to the trigger step.
This way you could just hard code the file being called on the trigger URL rather than a variable.
There could be a timing issue if multiple alerts come in though.
Right now the copy to folder happens after the trigger step is done, so we would not be able to call the copied file for the trigger url



Code:
[14.10.2020, 09:58:02.037]:             DetectObjects> 10.13.13.70:89 - (5/6) Performing alert actions:
[14.10.2020, 09:58:02.679]:     MergeImageAnnotations> Merged 2 detections in 656ms into image C:\Users\ADMINI~1\AppData\Local\Temp\2\Cam5.20201014_095800883.jpg
[14.10.2020, 09:58:02.684]:           CallTriggerURLs> -> trigger URL called: http://XX/admin?camera=Cam1&trigger&jpeg=E:\aiinput\Cam5.jpeg&user=XX&pw=XX, response: 'signal=green profile=1 lock=0 camera=Garage1HD_rtsp '
[14.10.2020, 09:58:02.767]:                   Trigger> Copying image to network folder...
[14.10.2020, 09:58:02.769]:                 CopyImage> File copying from C:\Users\ADMINI~1\AppData\Local\Temp\2\Cam5.20201014_095800883.jpg to E:\aiinput\Cam5.jpg
[14.10.2020, 09:58:02.771]:                   Trigger> -> Image copied to network folder.
 
It isn’t an option. When using 5.3.3 that is just want happens. But you need to ensure you have a pre trigger recording of a few seconds to ensure it can go back to get the image.


I'm on the latest 5.3.3.8, but the substream is not taking the Main-Stream's snapshots. It's still the substream low quality JPEGs.

1602696024985.png
 
Your pre trigger video buffers must be long enough to capture a full frame before it will work. Try setting it to 1 or 2 seconds


I just tried 2 seconds, and 5 seconds, still getting the low quality substream jepgs.

there's no box i need to tick to make this take Main-stream's snapshots?
 
I just tried 2 seconds, and 5 seconds, still getting the low quality substream jepgs.

there's no box i need to tick to make this take Main-stream's snapshots?
I tested it when the facility became available and it worked except occasionally I would get a corrupt image from my 12 Mp stream so have disabled it for the moment. No other changes were required in addition to the video buffer. Check out the help file, it has been updated to explain the new functionality. Perhaps it has got broken in the latest version!
 
Check under trigger tab. It has a high-res option I think. Can't remember the combo needed lol.

I think the downside to this though is it will duplicate the images. You will get one in the alerts folder (which is what it uses to send to devices for alerts) and one in the folder you have set for AI tools.

That said I don't have that checked because the above and my AI tools is getting high-res snaps. Your i-frame on your cameras needs to be 1.0 in BI and your pre-buffer can be 1.1 and that should be it honestly.
Doing the above checkbox makes the emailed alerts high-res too though but again, you end up with duplicate images in locations and thus more processing wasted.
 
Last edited:
I just tried 2 seconds, and 5 seconds, still getting the low quality substream jepgs.

there's no box i need to tick to make this take Main-stream's snapshots?
I have just tested the new functionality and it appears that it only works if your camera is sending H.264. When I change the stream to H.265 it stops working.
 
I have just tested the new functionality and it appears that it only works if your camera is sending H.264. When I change the stream to H.265 it stops working.
All my cameras are H265 and AI is getting high-res photos.

Make sure your sub and mainstream have the same key frame settings.
 
That interesting. I had tried the beta and it didn't work but just tried your suggestion deepquestai/deepstack:gpu and that works.

Speed doesn't seem any better though.

Edit- times do seem better but its processing a number of images then stopping.
I was able to install the GPU version on bare metal, however it will only process 1 image before stopping. Can you share your setup and process please?
 
Question for anyone that knows.

So I have my trigger and cancel alert URLs setup and working now.
I have my alerts set to re-trigger after 3 seconds unless active still and my image set to capture every 3.5 seconds.
The issue I have is sometimes the trigger image does not match the AI image because of delay processing or the fact that BI may only show one on the alert but it took 2 images.

For example BI takes snap, adds to thumbnail, while AI is processing another image is taken. The first image ends up as canceled because say a car JUST started into scene so motion was hit but not enough for AI to know it's a car. The second image knows it's a car so now it triggers. Well BI only made the one entry despite 2 images so when it triggers the email is the first image with maybe nothing in it of enough value. Funny part now is the alert is cleared from timeline due to the cancel but flagged cause I'm still using flags for trigger at the moment but this is because the same issue, multiple images but 1 entry in BI.

Anyway I know in the example you could just modify the motion to require being in the view more before trigger but that was just my example. It happens in other cases where you wouldn't want to do this so trying to understand if my times need to be adjusted or some way to make BI use all the images and not just the one etc.

What am I doing wrong or can I change to improve this? I only want multiple images in case the first misses. I may see if Ken can set an image time limit too (for times when it is constant motion I don't need non-stop alerts and maybe dynamic mask don't help since it can move a lot). Like mowing the yard and being lazy to not want to disable alerts. Well I can get hundreds of emails LOL. I only need 1 if it's a legit one and still part of the same alert but that's another issue.

Thanks!

Can you share what URL you are using to cancel the alert? I haven’t been able to get this working.


Sent from my iPhone using Tapatalk
 
All my cameras are H265 and AI is getting high-res photos.

Make sure your sub and mainstream have the same key frame settings.
Sadly it would appear that some cameras running H.265 do not work with this function. After your post I began to think it was pilot error on my side but after several hours of testing I have proven conclusively that the issue is to do with camera compatibility. For example I have amongst others two HikVision cameras (see attached) one works with switching H.265 streams the other will only work when set for H.264. I could post all the settings but you would simply see that they are all the same with the exception of camera stream resolutions.
 

Attachments

  • HikVision 1.jpg
    HikVision 1.jpg
    52.5 KB · Views: 41
  • HikVision 2.jpg
    HikVision 2.jpg
    52.9 KB · Views: 41
what are you guys doing with the SQL database file, after a while the file gets so big, when I RDP to my Blueiris machine to look at AI tool, it will try to initialize the history forever. IF i delete the file, it opens right away.

@Chris Dodge