My favorite part of this story is that you tried to track him with a GPS collar and he took it off in the middle of the road so you probably thought he was in peril or something Cats truly are just scheming lions that live in our house.
She was detecting cats in an indoor environment though, which is much more similar to the COCO dataset the model was trained on, and Tuco was a much larger part of the scene and well-defined in the image.
the difference there is that Micah didn't try the futile task of attaching thing to cat, rather Tuco Flyer moved camera to cat (ruclips.net/video/JU6omrP2iFU/видео.html)
Food for thought: I've been using a Tile (Mate), and a silicone case (ie: amazon: "Counlisha+tile+case"), and ESPresense to know when my dog comes back. I simply turn on an input boolean and there's a TTS notification when he comes near the front door. You could try with various ESPresense devices (fine tune them so there's no conflicts), around the house, near the windows, and you would - at least - know distances, and what side of the house the cat is roaming around.
Frigate's been doing a better job at detecting him in a door camera with the light colored decking, so it's just half of the doors that it can't detect him at.
lol after I set up my little frigate system I noticed there's a cat that pops through the area behind my building almost every night. Luckily said cat doesn't drop off any surprises for me.
For the AI door unlock idea, consider avoiding primary entrance for your homes security. Or combine it with some sort of Multifactor... Like a fingerprint can be forced, so can face recognition...
I have Frigate set up on a usff Dell 7010 with coral tpu using WYZE pan cam v2 I had many false positives. In car headlights a fire hydrant in my front yard got picked up as a person 5 to 6 times a night. Until I raised my threshold to 70% and my detection to 76 % did my person detection become almost flawless.
I have a friend with home cameras and an AI dongle for subjects recognition, it keeps sending him email alerts with photos of neighbour cats tagged as "person".
If the stream decoding for the detection is bad it can produce corrupted output. Then you end up with zebras. See if what the detector sees is actually proper video. It may be using a different stream for the actual snapshots and images vs detection making it very hard to be sure.
I haven't noticed more artifacts than my other cameras. The default bitrates are lower than I'd expect for this camera, so that could be related to the artifacting. Having to use the app to enable web ui and RTSP is irritating. But otherwise functionality is what I'd expect over the web UI + RTSP. I haven't tried the ONVIF integration with Home Assistant, but apparently that works well to translate on-camera detection into HA sensors.
Cheers, have you shared your frigate.yaml file? Im curious what hardware decoding or camere resoloutions you used as i have just got that camerz. Also you tried v14? Cheers, new subscriber
Thank you for posting this great tutorial. I was looking for it for quite a while. I am looking for a board to place inside my car. I want to use realtime object detection to recognise whenever a certain type of car is passing by in front of my car. And I want the board to notify me by 4g/lte sms( no wifi connection so i have to add a gsm/lte/4g module I am looking at 2 boards: 1. Raspberry pi 5 8Gb with Google Coral USB Accelerator 2. Google Coral Dev Board 4Gb Which one is the best option?
I would be cautious to continue buying Google Coral products in 2024, as Google hasn't updated most of their repos in 2 years and it's been like pulling teeth to get them to merge pull requests from the community (all of their people active on those repos seem to have been let go). You should be able to do it with all USB modules, so a Pi 4 would be adequate. The pre-trained models for Frigate only detect 'car' generically, not specific models, but you can take the generic detection and process the results through a identification neural network (which takes a known object and identifies the type/model/...) to pick out the exact car from the snapshot of 'car'. Viseron being more plugin based might be better at this task.
My older cameras support v6 and it's easy to compute their LLA from the MAC address sticker on the box, but now they are banned from import in the US so that's fun. I'm actually using v6 LLAs in my normal Frigate setup now and that works as well, other than the usual Docker v6 issues.
So, fun fact, these cameras do have a v6 EUI64 LLA and the RTSP port is open on it (but not HTTP/S) and it functions correctly. So once you do the initial setup you can use the LLA to actually use the cameras.
I have been working on cat detection for 5 yrs since I moved into my current house with all the neighbours cats using my backyard as a litter tray. Thankfully cat poo free for a long time now due to my cat detection AI and irrigation system which sprays them if they come close. But only due to BlueIris/Codeproject and a GPU. Frigate just did not do the job especially in low light conditions. great for people detection and low power consumption though
As far as sponsored videos go, this one was probably the best executed one I've watched. Congrats on the sponsorship too!
Thanks!
My favorite part of this story is that you tried to track him with a GPS collar and he took it off in the middle of the road so you probably thought he was in peril or something
Cats truly are just scheming lions that live in our house.
It was trying to get a car to run it over
Micah Elizabeth Scott did quite a bit of development on a cat tracking camera many years ago. As I recall, she got it working pretty well.
She was detecting cats in an indoor environment though, which is much more similar to the COCO dataset the model was trained on, and Tuco was a much larger part of the scene and well-defined in the image.
the difference there is that Micah didn't try the futile task of attaching thing to cat, rather Tuco Flyer moved camera to cat (ruclips.net/video/JU6omrP2iFU/видео.html)
I love your videos. Thank you for making them.
Cats: 1, AI: 0
Food for thought: I've been using a Tile (Mate), and a silicone case (ie: amazon: "Counlisha+tile+case"), and ESPresense to know when my dog comes back. I simply turn on an input boolean and there's a TTS notification when he comes near the front door.
You could try with various ESPresense devices (fine tune them so there's no conflicts), around the house, near the windows, and you would - at least - know distances, and what side of the house the cat is roaming around.
Frigate's been doing a better job at detecting him in a door camera with the light colored decking, so it's just half of the doors that it can't detect him at.
I watched this cause of kitty
He appreciates his acting job
lol after I set up my little frigate system I noticed there's a cat that pops through the area behind my building almost every night. Luckily said cat doesn't drop off any surprises for me.
No dead mice as peace offerings?
For the AI door unlock idea, consider avoiding primary entrance for your homes security. Or combine it with some sort of Multifactor... Like a fingerprint can be forced, so can face recognition...
Ooh, homelabbing, security and ev! What’s not to love.
Shout out to reolink. I love my Duo 2
He actually has a group of zebras in the neighbourhood ... but the camoflauge is actually working so well, you cant actually see them.
I have Frigate set up on a usff Dell 7010 with coral tpu using WYZE pan cam v2 I had many false positives. In car headlights a fire hydrant in my front yard got picked up as a person 5 to 6 times a night. Until I raised my threshold to 70% and my detection to 76 % did my person detection become almost flawless.
It definitely takes some tuning, but then it works very well (for people + cars)
So you are surveilling your cat with spy satellites? That must be a really secure cat. 😇😉😎
I have a friend with home cameras and an AI dongle for subjects recognition, it keeps sending him email alerts with photos of neighbour cats tagged as "person".
If the stream decoding for the detection is bad it can produce corrupted output. Then you end up with zebras.
See if what the detector sees is actually proper video. It may be using a different stream for the actual snapshots and images vs detection making it very hard to be sure.
It's caused by shadows in my case
Put a camera on the cat for PoV 😄
he would never tolerate that
PoV over PoE
I was really hoping you figured this out as I also failed to get frigate to detect my cat.
Blueiris, CPAI and a gpu is the only way to go for small animals outside. 4 yrs now and perfected it.
Supportive comment, since I do have neither animals nor children.
What's your experience with Reolink?
I used to have reolink and I got a lot of artifacts. I used Zoneminder, Shinobi, Motioneye and now Frigate.
I haven't noticed more artifacts than my other cameras. The default bitrates are lower than I'd expect for this camera, so that could be related to the artifacting. Having to use the app to enable web ui and RTSP is irritating. But otherwise functionality is what I'd expect over the web UI + RTSP. I haven't tried the ONVIF integration with Home Assistant, but apparently that works well to translate on-camera detection into HA sensors.
You should leave t he built in time on because adding it with Frigate requires processing
Cheers, have you shared your frigate.yaml file? Im curious what hardware decoding or camere resoloutions you used as i have just got that camerz. Also you tried v14? Cheers, new subscriber
Thank you for posting this great tutorial. I was looking for it for quite a while.
I am looking for a board to place inside my car. I want to use realtime object detection to recognise whenever a certain type of car is passing by in front of my car. And I want the board to notify me by 4g/lte sms( no wifi connection so i have to add a gsm/lte/4g module
I am looking at 2 boards:
1. Raspberry pi 5 8Gb with Google Coral USB Accelerator
2. Google Coral Dev Board 4Gb
Which one is the best option?
I would be cautious to continue buying Google Coral products in 2024, as Google hasn't updated most of their repos in 2 years and it's been like pulling teeth to get them to merge pull requests from the community (all of their people active on those repos seem to have been let go).
You should be able to do it with all USB modules, so a Pi 4 would be adequate. The pre-trained models for Frigate only detect 'car' generically, not specific models, but you can take the generic detection and process the results through a identification neural network (which takes a known object and identifies the type/model/...) to pick out the exact car from the snapshot of 'car'. Viseron being more plugin based might be better at this task.
@@apalrdsadventures thanks for the heads up about the Coral
When a new product comes already obsolete
My older cameras support v6 and it's easy to compute their LLA from the MAC address sticker on the box, but now they are banned from import in the US so that's fun.
I'm actually using v6 LLAs in my normal Frigate setup now and that works as well, other than the usual Docker v6 issues.
So, fun fact, these cameras do have a v6 EUI64 LLA and the RTSP port is open on it (but not HTTP/S) and it functions correctly. So once you do the initial setup you can use the LLA to actually use the cameras.
@@apalrdsadventures Good to know, but still... no full functionality on current standards
Cat videos. It was only a matter of time.
I have been working on cat detection for 5 yrs since I moved into my current house with all the neighbours cats using my backyard as a litter tray. Thankfully cat poo free for a long time now due to my cat detection AI and irrigation system which sprays them if they come close.
But only due to BlueIris/Codeproject and a GPU. Frigate just did not do the job especially in low light conditions. great for people detection and low power consumption though
Your condenser unit isn't level, I'm pretty sure that will damage it over time.
As long as it's mostly upright I don't think it matters too much.
shoulda said sherlock was also to scale lol 😂
Nice Chinese cyberweapon you're shilling....