[tool] [tutorial] Free AI Person Detection for Blue Iris
- Thread starter GentlePumpkin
- Start date
Cameraguy
Known around here
- Feb 15, 2017
- 1,502
- 1,159
Deepstack windows version was a bit quicker for me but I switched to docker version because it is saving me CPU
IAmATeaf
Known around here
That is impressive. My CPU is a i5-6500 and processing a 1080p image takes around 3.5 seconds. Am using docker desktop with WSL2 and DQ for docker.
Hi everyone,
please mind my english i am trying as good as i can. I am from Germany and its been ages since i had to use english.
after 20Hours of working on the System i have now finally got everything working. Tomorrow ill buy Blue Iris since the system is GODLIKE.
Let me tell you a bit about my Setup. And along those lines I will tell you where i had problems and how i fixed them.
First of all big thx to gentle Pumkin for the programm AITool. Even though I am using it at the moment I plan to change a few things here an there to fully integrate it in my smart home system. Things like mqtt and such.
Here my Setup:
Super Micro Server with 2 Intel Processors running VMWare ESXI 6.5 196GB of ECC Ram
On this Setup I have 14 Virtual Machines running. Mostly Debian systems like plex Owncloud MQTT Influx AI systems and others.
I set up a Win 10pro VM for Blue Iris gave it 16GB Ram and 2 Processors with 4 Cores each.
I have 8 Mobotix Thermal Cams arround the house connected to Blue iris.
The only issues i had was that AI tools didnt want to connect to the DeepStack AI.
What did i do to solve it?
well the only way i got it running was with docker and i had to use the noavx version. By my analysis there seems to be a problem with the windows version and the docker version at the Moment.
If there are any questions ill answer them as far as i can.
please mind my english i am trying as good as i can. I am from Germany and its been ages since i had to use english.
after 20Hours of working on the System i have now finally got everything working. Tomorrow ill buy Blue Iris since the system is GODLIKE.
Let me tell you a bit about my Setup. And along those lines I will tell you where i had problems and how i fixed them.
First of all big thx to gentle Pumkin for the programm AITool. Even though I am using it at the moment I plan to change a few things here an there to fully integrate it in my smart home system. Things like mqtt and such.
Here my Setup:
Super Micro Server with 2 Intel Processors running VMWare ESXI 6.5 196GB of ECC Ram
On this Setup I have 14 Virtual Machines running. Mostly Debian systems like plex Owncloud MQTT Influx AI systems and others.
I set up a Win 10pro VM for Blue Iris gave it 16GB Ram and 2 Processors with 4 Cores each.
I have 8 Mobotix Thermal Cams arround the house connected to Blue iris.
The only issues i had was that AI tools didnt want to connect to the DeepStack AI.
What did i do to solve it?
well the only way i got it running was with docker and i had to use the noavx version. By my analysis there seems to be a problem with the windows version and the docker version at the Moment.
If there are any questions ill answer them as far as i can.
There's definitely something unstable/missing with DeepStack's installation for Windows -- I don't know if there's a DLL that's not included with DeepStack's installation or what. When I google'd the specific pyd file that was erroring out (tensorflow) I found other posts un-related to DeepStack where people were having issues with it. Someone wrote a python script that you can run from Command Line and it scans the machine for dependencies for tensorflow and alerts you to what's missing. It came back with a handful of DLLs and things that needed to be installed. While most of it was freeware, there was 1 dependency for NVidia CuDNN which required a developer account to download and I didn't feel like jumping through all those hoops.
Can anyone elaborate on the URL for getting the memo into BI console? I used this URL but I'm getting literally '[summary]' showing up in the BI Console.
Attachments
Cameraguy
Known around here
- Feb 15, 2017
- 1,502
- 1,159
short name&trigger&memo=[summary]&user=username&pw=password
Cameraguy
Known around here
- Feb 15, 2017
- 1,502
- 1,159
must use 1.67 preview 7:
Releases · gentlepumpkin/bi-aidetection
Alarm system for Blue Iris based on Artificial Intellience. - gentlepumpkin/bi-aidetection
github.com
So the ONLY way to auto start things is with Win 10 Pro Docker? I did the AITool as a service to make it start with windows and want to make everything startup the same so I can set my little box somewhere and forget it without the worries of having to go back to it to click some programs. Is there a downside to using the Win 10 Pro Docker route?
D'oh - thank you! I had grabbed the newest version that was uploaded in the 1st post, but apparently it's a few versions behind.
Edit: That worked beautifully!
The best part about having a 5 year old kid is saying "Hey do you want to help dad test his cameras?! Go run a lap around the house!" And he does it with joy LOL
Are those scripts? I have no clue where to start with that or even where to put it :/
How did you make AITool as a service?
Same procedure. I use NSSM.
Also see this post
[tool] [tutorial] Free AI Person Detection for Blue Iris
The other process error is because BI hasn’t quite finished writing the file so has it locked open, if you look it retries and then succeeds. What is strange are try 2 separate requests to DQ, 1 using an IP address, the other using local host but both reference different ports. Do you have 2...
I do not have python or anything extra besides what I needed to get my BI, AITools and DQ working on a fresh Win 10 Pro install.How did you make AITool as a service?
Same procedure. I use NSSM.
Also see this post
[tool] [tutorial] Free AI Person Detection for Blue Iris
The other process error is because BI hasn’t quite finished writing the file so has it locked open, if you look it retries and then succeeds. What is strange are try 2 separate requests to DQ, 1 using an IP address, the other using local host but both reference different ports. Do you have 2...ipcamtalk.com
Well.. Deepstack is running on python so you have it.
All files is placed in the subfolders of the Deepstack folder.
So just follow the steps as you did for the AITools but just do those 3 as you posted before instead? I am stuck as what to do @ step 5. Sorry, I am not good with "coding" and follow step/step tutorials till I get the hang of things and this is confusing me where I put this information and or how to even do it. I understand how it was done for AITools but that is the extent of it :/
Last edited:
Thanks a thousand times for this awesome tool man, I am using it now for about a month and everyone I'm showing it to is completely stoked 
Running deepstack on a Pi4 with 4 GB ram and the Intel stick and it works like a charm.
I'm including it in my iobroker home automation to only record videos when both me am the wife are away which is working perfectly. Plus for example start an HD recording when the front gate opens even if there was no person detected.
Start all HD streams when the break in alarm fires etc...
Next step will be pushover notifications with alert images where your newest version with [file name] comes very handy!
One feature request/idea: how about a checkbox to select whether to send telegram alert images with the original image (as it is now) or including the detection boxes overlay? Maybe that's possible already but I wasn't able to find it.
Anyone having questions about the above setup just let me know
Edit: since I was wondering about this before: the Pi setup gives me results in less than 1 second. Even though I'm hitting it with images about 1MB size. Way faster than the 6-10 seconds I was getting with my non avx desktop PC.
Running deepstack on a Pi4 with 4 GB ram and the Intel stick and it works like a charm.
I'm including it in my iobroker home automation to only record videos when both me am the wife are away which is working perfectly. Plus for example start an HD recording when the front gate opens even if there was no person detected.
Start all HD streams when the break in alarm fires etc...
Next step will be pushover notifications with alert images where your newest version with [file name] comes very handy!
One feature request/idea: how about a checkbox to select whether to send telegram alert images with the original image (as it is now) or including the detection boxes overlay? Maybe that's possible already but I wasn't able to find it.
Anyone having questions about the above setup just let me know
Edit: since I was wondering about this before: the Pi setup gives me results in less than 1 second. Even though I'm hitting it with images about 1MB size. Way faster than the 6-10 seconds I was getting with my non avx desktop PC.
Last edited:
