yeah I was told tensorflow is baked into the container. Right now my issue is when the component tries to setup it sameta HA into a boot loop. Can’t figure out why, doesn’t throw any errors.
Once quick question - I have started writing my automation in Node-Red, and I was comparing what I have to your images above. In the “Current State Node” I cannot find how to change the “If State” to “Halt If State”.
Has this been changed in the palette or is there a way to change it?
Currently I am just using “If State is not 0”, but I wasn’t sure if this would have the same results.
UPDATE:
I think I found my own answer. They renamed it.
https://github.com/zachowj/node-red-contrib-home-assistant-websocket/pull/113
Thanks,
DeadEnd
I know this tutorial was intended for non hassio users but I was hoping to get help with setting it up via hassio.
I used this to get tensorflow installed. And it said it installed without error
https://github.com/hunterjm/hassio-addons/tree/master/tensorflow
But i get this error when home assistant boots
Unable to locate tensorflow models or label map
A few items have changed since I wrote this:
- I switched from vanilla HA docker to HASSIO, except running in a VM. I’ve found it to be much easier than trying to maintain a bunch of disparate docker containers.
- Node-RED has been updated and a few of the components have changed, as @DeadEnd had noticed.
- I switched from TensorFlow to Amazon Rekognition using this custom component by @robmarkcole. It works just as well and doesn’t require screwing around with TensorFlow files. Unfortunately it’s not in default, but the custom component works just fine.
If anyone is interested, I’d put together a different guide that will send you a camera stream when there’s an alert. It used the iOS application alerting.
Quick question… is this cloud processed?
It took a quick look at the repo, and from the details of the pricing, etc. it appears that your image is sent to Amazon’s system for processing - correct?
This is good for users on Pi I bet as it offloads the processing, but users who want local processing will probably stick with another alternative.
Thanks!
DeadEnd
Correct, it’s in AWS. The time for processing and response is about on par with local TensorFlow. I’d love to have it local, but got tired of dealing with TensorFlow nuances. I’d switch back to TensorFlow is the AWS component became unmaintained or stopped working though. Not a lot of other options.
…and it looks like the Pushover custom component has taken a shit with the latest release. I’ve just updated my flows to use a function node and the latest Node-RED Pushover module, which supports image attachments. I’ll update the instructions this weekend. I really wish this would get merged.
I’ve posted an updated guide. The logic for the nodes has been updated for the current Node-RED version too.
Thanks TC. Can’t wait to dig into it.
Hi,
thanks for creating this great tutorial. I just got one question: Is tensor flow only able to tell me that it actually detected a person or can it also learn WHICH person is on the picture?
Only that a person is detected. If you want to know which person you need facebox
1 Like
Cool there is even an Home Assistant integration.
Sadly I am sure that this is 120% illegal in germany. We are not even allowed to have a slight view of any public place in the picture (i.e. sidewalk). There are cases where people do break into a house and after being sued, they sue the house owner because he filmed where he shouldn’t and it even happened in a case that the penalty for the house owner was higher than for the thief, lol.
I think creating a database with biometrical information about people faces will bring even more trouble 
are you able to black out an area?
I use this on my back garden as I don’t want to see into my neighbours’ garden.
The facebox thing requires you to teach it faces to recognise, so apart from the “recording a public area” you should be fine
Black out picture areas: That doesn’t matter. Even if it is not a real but fake camera but people in the public “feel” watched, the court will decide against you here.
Regarding facebox, I will essentially create a biometrical database of people. To be legal with this, everybody in that database would need to sign me a disclaimer
Germany is extremely strict in protecting personal data what is great on the one hand and makes your live unnecessary hard sometime on the other hand.
Also, while I am not allowed to protect my ground using cameras as I wish, I am filmed practically ANYWHERE else in the public (train stations, public places, stores, everywhere) and they even plan to put face recognition in there…
Well have fun 
1 Like
Will just stick to object detection