Hi all-
Just curious of those who have Blue Iris and use the MQTT feature to integrate with Hass.
I currently have my front door camera set to send an MQTT message when there’s motion as determined by BI. I have a Hass automation that fires an HTTP command to blue iris to snatch the current still from that camera and then alerts me via the Hass app when I’m at home. I also have an SMS sent to my phone directly from BI for those times I’m not at home (I don’t keep a persistent VPN or poke holes through the firewall). Here’s the issue: the Hass app version doesn’t capture the thing that set off the alert because I think it’s too long (between BI MQTT message and screen grab - i think it sends the MQTT message AFTER the motion is complete as opposed to when it first sees it maybe) and the UPS guy or whoever has left. The direct from BI text message though gets it perfect.
So, i’m curious how anyone here does this. I suppose I could have the MQTT message send me the alert directory and filename of its alert picture (which is basically what’s sent via text) and have Hass get it via windows share or something. Another way is for BI to FTP the image to Hass, but I don’t know how to get an FTP server easily on it.
I use BlueIris with the DOODS integration. This has the added advantage of filtering out most false alarms, since it is only looking for people. When MQTT triggers I run the image processing 3 times with 1 second delay in-between to deal with the issue you described.
Other than that you will need to probably play with your camera properties, object detection settings, etc. With the added DOODS filtering, you can be aggressive with these knowing that it will filter out most falses.
But it is hard to catch those UPS guys they get faster everyday. I have also noticed they like to throw things more when they know you have a camera, trying to dodge it.
What’s DOODS? Is that the AI object detection? If so, I tried for like 1 week straight and couldn’t get the AI part to run on anything I owned. It just wouldn’t run correctly. It seem to install ok, the http link worked enough to say “up” but it refused to process any pics. I sooo wanted it to work too.
Yeah it’s one of them, there are several. I used the docker container so it pretty simple, I don’t remember doing any tuning on it, but I am running on vmware/linux.
I did try tensorflow previously standalone and it was a lot more trouble to get working.
I tried it in a docker container on Hass, but the reason it wouldn’t run is because I run Hass in a VM, I run BI in a VM, I run MythTV on a VM, etc. I tried it on the “bare metal” / Windows 10 but it would run on that either.
Also, which one did you use? I had no success with DeepStack AI. It just wouldn’t process images no matter what. Ideally, I’d like one I could run locally like DeepStack.
There is no http link for DOODS. Sounds like you might mean watsor. I have been trying that one and can’t get it going yet either. I haven’t had much time to play with it yet though.
I have tried many of them including watsor, deepstack and DOODS and I keep going back to DOODS because it works well for me.
I use blue iris and HA, but not mqtt. I do what you are trying to do with a couple camera components. Depends on what type of cameras you have though. I have dahua which work with the amcrest component and some hikvision that have their own component. Use the camera to detect motion and send that to HA via the component and then trigger the image processing scan from that. Works pretty well.
Do you have a link to how to get this running with BI? I never knew about this one. I wonder if it will run in my VM Hass in docker.
So does BI send mqtt messages when it detects motion every one sec? So how do I send or grab the images to get it into DOODS? Is it the HTTP screen grab command for BI? (http://ipadress/image/camshortname?q=quality&s=%size&user=xxxs&pw=xxx). Or somethings else? Then how does DOODS confirm that it’s a “yes” for detection to get BI to record or alert?
It seems that your desired outcome has changed as this thread has gone on.
You originally wanted the screen shot sent immediately but it was too delayed and missed the object. Now you seem to want object detection alerts.
If you still want to figure out how to get a good screen shot of the motion detection then it would help if you posted the current automation that you’re using to get all of this done. Maybe the timing issue could be solved there.
alias: Notify of front door activity
description:
trigger:
- entity_id: sensor.frontdoorcam
from: '*'
platform: state
to: activity
condition:
- after: 07:45:00
before: '23:59:00'
condition: time
action:
- data:
data:
image: http://<ip>0/image/Door?q=50&s=30&user=xxxx&pw=xxxx
ledColor: white
message: Front Door Activity
service: notify.mobile_app_sm_n960u