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

Awesome, that did the trick, thanks all!

Is there any way to tweak settings so that AI Tool is more likely to trigger BI? My tolerances are already 0-100% on AI Tool, yet there are plenty of images it scans where a person is in the picture but it fails to see the person and incorrectly calls it a false alert.
 
I can't get this to work no matter what. I've tried FOR DAYS. The AI tool just won't acknowledge the exisitence of the deepstack AI server. I tried a windows install on another machine, doesn't see it. Yes I can access it via URL and it says activated. Then I tried running both of those in admin mode, nope. Then I moved to a portainer container in Hass.IO like the video. Looks to be running (deepstack) - but AI Tool refuses to see it. Tried with all firewalls off - nope. I'm about to throw this in the lake.
 
I can't get this to work no matter what. I've tried FOR DAYS. The AI tool just won't acknowledge the exisitence of the deepstack AI server. I tried a windows install on another machine, doesn't see it. Yes I can access it via URL and it says activated. Then I tried running both of those in admin mode, nope. Then I moved to a portainer container in Hass.IO like the video. Looks to be running (deepstack) - but AI Tool refuses to see it. Tried with all firewalls off - nope. I'm about to throw this in the lake.

Try the Docker container approach. Same BlueIris config, different trigger manager. Comes with the Deepstack AI server as part of the package and pre-configured so there's no connectivity issues to worry about. Install it on the same machine that's running BlueIris for easy setup.
 
Try the Docker container approach. Same BlueIris config, different trigger manager. Comes with the Deepstack AI server as part of the package and pre-configured so there's no connectivity issues to worry about. Install it on the same machine that's running BlueIris for easy setup.
I just read through it and was having a hard time understanding it. So it's basically both tools in one right? How do I edit stuff in docker? I've never used docker, but have used it in the portainer container in hass.io recently. I'm wondering if I could load it on that?
 
I can't get this to work no matter what. I've tried FOR DAYS. The AI tool just won't acknowledge the exisitence of the deepstack AI server. I tried a windows install on another machine, doesn't see it. Yes I can access it via URL and it says activated. Then I tried running both of those in admin mode, nope. Then I moved to a portainer container in Hass.IO like the video. Looks to be running (deepstack) - but AI Tool refuses to see it. Tried with all firewalls off - nope. I'm about to throw this in the lake.
You did select the detection API in Deepstack and you did select the right port? Please write me a private message with your Log, a screenshot of the deepstack website, a screenshot only of the url of Blue Iris Webinterface UI3 and a screenshot of your settings tab in AI Tool.
 
  • Like
Reactions: neile
I just read through it and was having a hard time understanding it. So it's basically both tools in one right? How do I edit stuff in docker? I've never used docker, but have used it in the portainer container in hass.io recently. I'm wondering if I could load it on that?

I have same issue I am trying to make it work in HomeAssistant.
 
Yes, it's both tools in one and pre-configured to talk to each other. Docker really isn't hard. Skip portainer and hass.io, it's just way more complicated than you need. I assume you have a Windows machine already running BlueIris so install directly on that.

Follow the quick start instructions I provided above. You only make two edits to the docker-compose.yaml file, which is plain text and can be edited in any text editor you like (I recommend VSCode, but Notepad++ or even Notepad works). You'll have the system up and running but not really processing anything. Then edit the sample triggers.json file to configure your triggers (similar to the configuration you've done in the AI Tool) and it'll start detecting things.

If you have additional questions about trying this route I suggest adding them to the dedicated thread for the alternate approach.
 
Last edited:
I can't get this to work no matter what. I've tried FOR DAYS. The AI tool just won't acknowledge the exisitence of the deepstack AI server. I tried a windows install on another machine, doesn't see it. Yes I can access it via URL and it says activated. Then I tried running both of those in admin mode, nope. Then I moved to a portainer container in Hass.IO like the video. Looks to be running (deepstack) - but AI Tool refuses to see it. Tried with all firewalls off - nope. I'm about to throw this in the lake.
Please see post # 338 on this thread. It goes over the prerequisites for Deepstack on Windows. Also, does your processor support AVX? If not you'll need to run Deepstack inside of Docker on a Linux machine using the :noavx command.
 
Please see post # 338 on this thread. It goes over the prerequisites for Deepstack on Windows. Also, does your processor support AVX? If not you'll need to run Deepstack inside of Docker on a Linux machine using the :noavx command.
No sure how to check this (AVX), but I did do your suggested patches and it made no difference - AI tool did not reach the server. Also, I ran it on a different server. I'm running out of CPU cycles and space on my VM where Blue Iris runs so I tried to run it on the host windows 10 machine. So maybe I shall put it all in the same Vmachine and try again. I am somewhat interested in nelie's new way, but being a lazy non-programmer the AI tool makes config 10x faster and clearer than editing the files (which I'll probably end up doing if this doesn't work). The VM is probably even more muddy or worse so far as AVX support. I'll try anyway.
 
If you install VSCode and edit the files with it it's not really any different than the AI tool. You get full auto-complete for all the properties, with tooltips to tell you what everything is. The sample file I provide has a basic setup specifically for talking to BlueIris: all you have to do is change the names of the cameras and copy paste the block a few times for each camera you want to connect to.
 
No sure how to check this (AVX), but I did do your suggested patches and it made no difference - AI tool did not reach the server. Also, I ran it on a different server. I'm running out of CPU cycles and space on my VM where Blue Iris runs so I tried to run it on the host windows 10 machine. So maybe I shall put it all in the same Vmachine and try again. I am somewhat interested in nelie's new way, but being a lazy non-programmer the AI tool makes config 10x faster and clearer than editing the files (which I'll probably end up doing if this doesn't work). The VM is probably even more muddy or worse so far as AVX support. I'll try anyway.
More modern CPU's support AVX. I have a HP Z420 with a Xeon E5 2960 V2 CPU that supports AVX, but like you I was oversubscribing my CPU as I have BI with 15 cameras and running Deepstack on it was too much. The easiest way to be sure aboout AVX support is to spin up a Ubuntu VM and do docker pull deepquestai/deepstack:noavx and then start it with the noavx tag and see if AI Tool can contact it. I ended up reviving an old Dell PowerEdge to run ESXi on with a Ubuntu VM for Deepstack. noavx is slower, but I still get response times under 2 seconds running on low mode.
 
Are there any plans to provide Deepstack AI with a list of known faces and only trigger action for unknown person event?
 
This is great progress!! I am really keen to see the implementation of @BrutusBoots and @GentlePumpkin to see the % and also to have the alerts 'flagged' for positive detections. Once this is released, I'll be ready to take the plunge and do the switch!