I purchased my LPR camera on one of Empire Andy's Black Friday sales. I knew the install was going to be a ton of work - and it was - so I had been putting off this project for a while. I was inspired by crc2004's post some time back. To install, I ended up running CAT6 direct burial cable with over 100' of hand trenching with an edge cutter tool. Many people advise against direct burial cable, but conduit would have been at least double the work. I have an invisible fence, which is about 2" deep and it's been working fine for years, so I'm hoping to have equally good luck with this cable.
Here's a picture close up and another from the road. I painted it with ultra flat camo paint.
Now, I'll fiddle with APLR software. I have a hand crafted system in my driveway doing APLR, which is using the SightHound API. I can stay under my API limits quite easily with the cars in my driveway, but I'd likely hit the limit if I tried the same for the road. My home built system has my driveway cameras ftp snapshots to a Jetson nano. If the nano finds a vehicle, then it crops and uploads the image to SightHound API. If the plate is recognized then my HomeAssistant will take appropriate action.
But the use case for the road is quite different, so I thought I'd try the commercial offerings. I've installed the Rekor Scout agent on a spare NUC. The home version requires you to use the cloud for storage. The cloud site seems to have limited options, to say ignore certain known plates, or to be able to add information to a plate. Or to alert when a new vehicle is in the neighborhood, etc..
The CPU on my NUC couldn't keep up at 4MP 15 FPS. I could do 4MP 5 FPS. They recommend 1080p. So I'm doing 1080p 20FPS which seems to be working OK. Fortunately, the IPC-B54IR-Z4E allows for two sub-streams, and I can configure the second substream to 1080P. Oddly, Rekor is reporting constant motion, and seems to be using 50% CPU non-stop. My home grown system is far more efficient, because the snapshots are saved based on IVS rules instead of constant processing, but I'd need to run my own ALPR service and build out a database.
Plate Recognizer seems a lot more expensive, unless you can fit inside their free plan. I know others are using BlueIris w/ DeepStack. but it seems you have to build your own database. Ideally, I could run everything locally and not have to pay a monthly fee.
Here's a picture close up and another from the road. I painted it with ultra flat camo paint.
Now, I'll fiddle with APLR software. I have a hand crafted system in my driveway doing APLR, which is using the SightHound API. I can stay under my API limits quite easily with the cars in my driveway, but I'd likely hit the limit if I tried the same for the road. My home built system has my driveway cameras ftp snapshots to a Jetson nano. If the nano finds a vehicle, then it crops and uploads the image to SightHound API. If the plate is recognized then my HomeAssistant will take appropriate action.
But the use case for the road is quite different, so I thought I'd try the commercial offerings. I've installed the Rekor Scout agent on a spare NUC. The home version requires you to use the cloud for storage. The cloud site seems to have limited options, to say ignore certain known plates, or to be able to add information to a plate. Or to alert when a new vehicle is in the neighborhood, etc..
The CPU on my NUC couldn't keep up at 4MP 15 FPS. I could do 4MP 5 FPS. They recommend 1080p. So I'm doing 1080p 20FPS which seems to be working OK. Fortunately, the IPC-B54IR-Z4E allows for two sub-streams, and I can configure the second substream to 1080P. Oddly, Rekor is reporting constant motion, and seems to be using 50% CPU non-stop. My home grown system is far more efficient, because the snapshots are saved based on IVS rules instead of constant processing, but I'd need to run my own ALPR service and build out a database.
Plate Recognizer seems a lot more expensive, unless you can fit inside their free plan. I know others are using BlueIris w/ DeepStack. but it seems you have to build your own database. Ideally, I could run everything locally and not have to pay a monthly fee.
