Just to add my two cents to this with some real-world data and as I just started playing with plate recognizer for the alert integration features. I have been running the OpenALPR local process from github since 4/8/2019 (thanks
@fernandw OpenALPR - A How-To on Ubuntu 16.04 Desktop). It runs on a separate PC, 100% local without any internet access needed and processes the plates from the camera's stream directly, so BI could be shut down and it's still chugging along. I have several scripts shared in the post above that scrape the ALPR processes output and log them to a MySQL database, and save an image capture of the plate to a folder. I added to those scripts to also post the plate to an MQTT topic so that Home Assistant can alert me to certain plates. I have two cameras on my street, one for east and one for west. We are a two-plate state but a lot of people don't put their front plate on properly, or their back plate is blocked by a trailer hitch accessory, etc. I am using the 12ZE cameras and experience the fail to focus at night issue sometimes, so two cameras also increases my odds that one is in focus!
Since installation, my database has 909,572 records in it. That includes every partial attempt to read a plate at all confidence levels (so if ABC123 goes by and it starts reading A8C123, ABCI23, etc.. those are all logged), If I filter by a Confidence of 85 or above, I get 779,510 results. Here is a monthly count over the last year or so of any confidence level.
Clearly 2,500 calls to Plate Recognizer are not nearly enough for my case, especially when you consider that each image sent is counted with them. At minimum I would have to pay the $50/month rate for 50,000 lookups, but would start missing some plates on busier months, that pushes me to the $150/month plan. WAY too expensive for a homeowner who does this more for curiosity than anything. My ALPR database only consists of images that returned a plate (or lettering off of the side of a truck). In the short time I've had Plate Recognizer configured I have had four cars go by. The camera AI settings page says I have used 58/2500 calls since there is a breeze blowing the trees over the street causing shadow triggers. None of my alerts, nor my status log, show any plate reads from Plate Recognizer.
My requirement would be 100% local with a local SQL database. I took
@fernandw's web interface and customized if further to allow me to search by plate, date/time, camera, etc. I also made a database of neighbors and plates I want alerts on so I can filter by those. I also can break down counts by hour of the day, day of the week, etc. I want to do a lot more with it but just never took the time to do it.
Long story short, there are some real capture stats to compare for monthly needs. My street is a typical residential street, not a busy thoroughfare or cut-through to avoid anything. When I first got this up and running I took a drive around to see if I could place where all the passing cars were from and I could find most parked in neighbors driveways on my street and up to a few streets over.
