Deer alert, dog dispatched

[brianegge]

I’ve been progressively improving my system to detect when deer visit my property. As cute as they are I don’t like them eating my fruits and vegetables.
So my setup will send an alert to my phone, showing a cropped image, but it also is assigned to a unique tone. My dog has learned the tone means we have a deer and he gets excited to go out and give it chase. In the video below you can see I let the dog out about 40 seconds after the deer arrive. The video is cropped to 1/4 of the original 4K frame. I hope to replace the Lorex cam with a 5442 very soon. The shutter speed on the Lorex is too slow to catch a good image of the dog in motion.

 

[MrSurly]

Now, that's cool. I can foresee your next innovation will be the automatic operation of the doggy door, based on the deer detection system. way cool

 

[brianegge]

Now, that's cool. I can foresee your next innovation will be the automatic operation of the doggy door, based on the deer detection system. way cool

Exactly. I have the garage door automated to let the cat in, but I don't yet have a doggie door on the house.

 

[concord]

Do you have an invisible fence for the dog? Seems like your dog stayed on the lawn and didn't pursue into the woods.

 

[brianegge]

Do you have an invisible fence for the dog? Seems like your dog stayed on the lawn and didn't pursue into the woods.

Haha! Yes, he knows his boundaries! Many people in my area put up 6' steel fences to keep the deer out with a cattle guard at the front of the driveway.

 

[bmorgan]

Can I glom your deer model? Not sure I'm up to the task and my confidence in the bird model is low.

 

[brianegge]

Can I glom your deer model? Not sure I'm up to the task and my confidence in the bird model is low.
View attachment 80310

Are you using deepstack? Do you want a trained model or my dataset.
I only alert if the confidence is > .85-.90, but do object tracking > .4.
Ideally I’d have a one click way of confirming detections, which would add to the validation set or saying it’s a miss, which would add to the null set.

 

[bmorgan]

Yes. Deepstack, noob. Not sure which is best for me. Whatever is easiest.

Edit: looks like the trained model is my speed. I just stick the .pt file somewhere and add --MODELSTORE-DETECTION "C:/path-to-detection-models" to my startup?
I added the license plate sample as per above, but see no mention of it in AI Tool (vorlon).

 

[nbstl68]

Funny, it looks like the video flashed "dog" for the deer and "person" for the dog.

 

[brianegge]

I'll try training DeepStack this week.

Funny, it looks like the video flashed "dog" for the deer and "person" for the dog.

Yes, but an object tracker can easily clean that up. You start tracking an object and keep the class which has showed the highest probability over any frame. The dog was running so fast it was basically just a blur on my Lorex cameras.