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

There could be more but here is a list of all I'm aware of
person, bicycle, car, motorcycle, airplane, bus, train, truck, boat, traffic light, fire hydrant, stop_sign, parking meter, bench, bird, cat, dog, horse, sheep, cow, elephant, bear, zebra, giraffe, backpack, umbrella, handbag, tie, suitcase, frisbee, skis, snowboard, sports ball, kite, baseball bat, baseball glove, skateboard, surfboard, tennis racket, bottle, wine glass, cup, fork, knife, spoon, bowl, banana, apple, sandwich, orange, broccoli, carrot, hot dog, pizza, donut, cake, chair, couch, potted plant, bed, dining table, toilet, tv, laptop, mouse, remote, keyboard, cell phone, microwave, oven, toaster, sink, refrigerator, book, clock, vase, scissors, teddy bear, hair dryer, toothbrush.
Thanks @Brendon06, it looks like none of the items I'd like it to detect are listed. Hopefully they'll add news ones over time.
 
GPU performance on High

1612517597417.png

GPU performance on Low

1612517762133.png

Virtual Win10 for BI
Virtual Ubuntu 20.04 for DeepStack with 1080 (not Ti) passthrough
 
Should work with a 1030. Performance might be different.

Not sure how to get it going on Windows - as mentioned I have DeepStack running on Ubuntu and only BI running on Widows
 
i followed the instructions for windows installation of the gpu version
installed all the drivers for nvidia including creating the developer accounts to dowload everything
works pretty much the same as without hardware gpu acceleration for me (250-300ms) on medium
 
i followed the instructions for windows installation of the gpu version
installed all the drivers for nvidia including creating the developer accounts to dowload everything
works pretty much the same as without hardware gpu acceleration for me (250-300ms) on medium
I have like a $10 GTX 745 and it's 3-5 times faster for me. Check task manager to make sure it's actually using the GPU.
Should also note, when I was using the CPU I was having Bi reduce to 1080p, now I just leave them native 4k, as I like the high quality snapshots (I have AI Tool archive and send out verified snaps)
I am running 6 deepstack GPU instances as each uses <2% of my GPU
For me the beta didn't work either, downloaded the one in this forum. I checked the deepstack doco and they released a new beta, I am reluctant to try that, given it's been working great.
Anyone tried the latest beta?
 
I have like a $10 GTX 745 and it's 3-5 times faster for me. Check task manager to make sure it's actually using the GPU.
Should also note, when I was using the CPU I was having Bi reduce to 1080p, now I just leave them native 4k, as I like the high quality snapshots (I have AI Tool archive and send out verified snaps)
I am running 6 deepstack GPU instances as each uses <2% of my GPU
For me the beta didn't work either, downloaded the one in this forum. I checked the deepstack doco and they released a new beta, I am reluctant to try that, given it's been working great.
Anyone tried the latest beta?
you run it on windows or linux ?
 
[GIN] 2021/02/05 - 20:59:29 | 200 | 70.9997ms | 127.0.0.1 | POST /v1/vision/detection
found the same beta21.01 version on the forum. it seems to work now.

is there any plan to add minimum object size to aitool ?
i have several cats that are detected as 'human' in certain positions in the yard :)
 
i followed the instructions for windows installation of the gpu version
installed all the drivers for nvidia including creating the developer accounts to dowload everything
works pretty much the same as without hardware gpu acceleration for me (250-300ms) on medium

I had this same problem. You need to go to the deepstack.cc forums and download the beta release. It dropped my times from 350-400 ms to 70-80 ms. Let me see if I can find a link. pow...

 
  • Like
Reactions: Saargebeat
How is this detected as a person?:)
Minimum object size should filter those detections.
 

Attachments

  • person.png
    person.png
    1.3 MB · Views: 75
Hello Friends of AI

can someone explain how to setup dynamic masks? I like to get rid of detection of parked cars. I thought i did everything but it doesnt create any dynamic masks

I dont have static masks or anything.



Also i have 150-200ms response time with RTX3070
Using newest beta

I tried MODE High but then i got more than 300ms

Also CPU goes up to 100% starting with 13% but GPU only quite stable 1% - during processing maybe 5%


Maybe a bit OT: I get the 10sec Push Images, but most of the time they start to late (person is gone already). Any idea how to improve?
 
Last edited:
Hello Friends of AI

can someone explain how to setup dynamic masks? I like to get rid of detection of parked cars. I thought i did everything but it doesnt create any dynamic masks

I dont have static masks or anything.



Also i have 150-200ms response time with RTX3070
Using newest beta

I tried MODE High but then i got more than 300ms

Also CPU goes up to 100% starting with 13% but GPU only quite stable 1% - during processing maybe 5%


Maybe a bit OT: I get the 10sec Push Images, but most of the time they start to late (person is gone already). Any idea how to improve?
I literally just clicked the "enable dynamic mask" button on each camera. Try a different version of AI Tool, perhaps your version has a bug.
Pre-recording should stop you missing any footage, but I recommend recording 24x7 continuous and have AI Tool flag footage for you.
I don't recall having any dynamic masking issues, I am currently running a self compiled version.
 
I literally just clicked the "enable dynamic mask" button on each camera. Try a different version of AI Tool, perhaps your version has a bug.
Pre-recording should stop you missing any footage, but I recommend recording 24x7 continuous and have AI Tool flag footage for you.
I don't recall having any dynamic masking issues, I am currently running a self compiled version.


I took the lastest versions of AITool and the Fork.
Maybe i can get a confirmation of somebody running the same software versions?! Or any non-self-compiled one :)

Also maybe there is a short manual how it should work. Whatever i try, i dont get any dynamic mask
 
If there are 4 alerts in a row, separated by 2 seconds each, only the first alert is being sent to Telegram, for the other 3 alerts it fails with ERROR sending image to Telegram error, no further details. Chat id and telegram token are correct since at least one image is sent to telegram. Cooldown is set to 0.033. Should I just try different values for caption and/or filename or any other parameter? Does someone have the "magic" settings that really work?
FYI, I migrated to vorloncd from the original aitool. Aitool used to to be able to send all the images to telegram.
 
If there are 4 alerts in a row, separated by 2 seconds each, only the first alert is being sent to Telegram, for the other 3 alerts it fails with ERROR sending image to Telegram error, no further details. Chat id and telegram token are correct since at least one image is sent to telegram. Cooldown is set to 0.033. Should I just try different values for caption and/or filename or any other parameter? Does someone have the "magic" settings that really work?
FYI, I migrated to vorloncd from the original aitool. Aitool used to to be able to send all the images to telegram.

After playing a while with the settings, I managed to fix the issue myself, therefore the ERROR sending image to Telegram error no longer shows up if check the Queue actions setting:
1612775791463.png
 
I took the lastest versions of AITool and the Fork.
Maybe i can get a confirmation of somebody running the same software versions?! Or any non-self-compiled one :)

Also maybe there is a short manual how it should work. Whatever i try, i dont get any dynamic mask

I see no one has replied, if you don't want to compile yourself, you may wish to try this version to deal with the masking issue.

Dynamic masking - nothing created · Issue #77 · VorlonCD/bi-aidetection - Dynamic masking - nothing created #77

 
  • Like
Reactions: Saargebeat
I have the cameras and the Server time synced with nettime

Everytime I look at Aitools history the time on the first motion event jpg is about 4 seconds behind the aitools history listed time, every jpg after the first are all in the same second.

The jpg time is the same second as the file timestamp

AMD 3700
64GB ram
NVME for OS and aitools and blueiris
SSD for jpgs
Hardwired camera
Nvidia 1660 super

deepstack GPU 2021.02.1
same issue with previous two betas as well

AiTools 2.0.627.7696
but it's been the same time gap with the previous versions as well

Any ideas why the time gap or how I can improve it?
 
y'all, i did it! 157 pages read. It took a month (did it when I had a sec here and there), and I'm not sure it was worth it (over a more targeted approach that is) but I def have a much better understanding of how everything works together and where the project has gone. Little embarrassed about some earlier questions (thanks for your patience @Village Guy and @austwhite !). I put links to the posts of everything I thought might be useful into an email draft as I went, and I'm currently making a 2nd pass but I did have 1 clarification before i get too far:

Is it required to be continuously recording on SOME stream (Main stream with flags or sub stream with triggers) in order for this to work? Or is it possible to simply have jpegs taken on motion without any continuous recording and then sent to AI Tool for confirmation and triggering?
 
  • Like
Reactions: Nierka