Gladys Assistant 4.48: Proactive AI arrives in Gladys Assistant!

News
21

With a little distance you could almost see the license plate; he should be able to read it when he comes back.

22

The problem is the timing.
Because we’re talking about a static capture of the video stream…

So when I arrive home, I open the gate > trigger the scene.
But I don’t know exactly when Gladys will hand the image to ChatGPT for analysis…
A little too early and we won’t even see the car / a little too late and the license plate will be too low in the frame. :confused:

23

nothing prevents you from taking a series of 3 or 4 images at 0.5 s intervals or more depending on what seems appropriate to you

then ask it whether it sees the license plate

and if it sees the

24

You put a fluorescent strip on your dashboard visible to the camera.

1 Like
25

:joy: I never thought I’d get into tuning

1 Like
26

Indeed, it’s a solution… But it’s a bit of a hack, I think.

I’ll test it anyway and I’ll get back to tell you! :+1:

27

Hello everyone,

The ChatGPT integration into scenes works like a charm, thank you very much! :clap:
On the other hand, does anyone have an example of a prompt that makes ChatGPT not automatically reply?
Example: if I monitor my camera in front of my house every minute and I ask it: « Warn me if you see a person in the yard in front of the gate, otherwise there’s no need to notify me. »
Well, I still systematically get a response along the lines of: « I don’t see anyone in the yard in front of the gate. » (with some variations in each message, but the gist remains the same).
That said it works well because when there actually is someone: « There is someone in the yard in front of the gate. » :smiley:

28

Have you tried using only the beginning of the prompt?

« Let me know if you see someone in the courtyard in front of the gate »

29

Yes, and even with nuances like « Only if », « If ever »,… same result: he’s chatty! :grin:

30

Maybe an idea, but I haven’t tested it: you write in the prompt: "is there someone in front of the gate? Answer me only with 'yes

31

Great :smiling_face:

That’s not possible right now, it always gives a response! But it could be added :grin: I’ll look into it

3 Likes
32

Thanks everyone, before trying the analysis again I’ll first limit the number of analyses. I’ll try to capture the ONVIF events from my camera in Gladys. :wink:

2 Likes
33

Hello.

Indeed the problem is being approached, IMHO, the wrong way: you need to trigger on an ONVIF event.
Personally, I use Onvifeyes and Hooks that set MQTT values which Gladys then interprets… In the absence of event handling on Gladys’ side :smiling_face:

Have a good Friday everyone,
Jean

1 Like
34

I’ve just added an option so that ChatGPT doesn’t reply if there’s no need to reply :slight_smile:

This will be included in the next version of Gladys:

1 Like
35

Gladys Assistant has been added to Selfh! :tada:

1 Like
36

Yes, he told me he’d add it :slight_smile: So great!

3 Likes
37

So cool! Curious to know if you’re going to see any feedback on this post!! :+1:

38

Agreed! :blush:
So, I completely rethought everything and went with a Frigate/MQTT solution to be compatible with other future cameras that might not be ONVIF-compatible. :wink: The only downside is that my Raspberry Pi has to do the person detection, whereas the camera does it natively.

39

How fast!:face_without_mouth: thanks :blush:

40

A question following up on the discussions of @guim31 and @Prof_Techno: when an image is attached to a request to the AI. Is it a capture taken when the scene is triggered or is it an image taken with the latency specified in the « camera » integration?

1 Like