This is one of those situations where you have to know the answers before you ask the questions. Most LPR companies would tell you that they read license plates in the high 90% range. Fair Enough. However that’s of license plates that can be read.
What about plates behind a bike rack, or trailer hitch, or covered with mud or snow or ice? What about cars with no plates, or dealer plates in the rear window? What about plates from a foreign country with strange characters or fonts?
I remember a few years ago talking to an LPR company rep in the UK. I asked him what the percentage read rate was. He asked me where was he doing the reading. I said in the UK. He said better than 99%, a requirement by the government to provide LPR technology to the police and enforcement agencies. I then said, “In the US.” When he stopped laughing, he said, “Maybe in the mid 80%, on a good day.”
Seems that in the UK, all license are the same – that is white on black, and the same font. In the US, every state is different, plus in most states, there are different designs, pictures, and PR slogans. I understand there are more than 50 different designs in California alone. The computer must be trained to differentiate between all these and do it accurately.
Technology has moved on since my conversation in the UK. However even if the camera/computer combination is capable of reading and decoding 100% of plates it can see, what about those listed in paragraph above?
If I am scanning cars on the freeway, or those parked on a city street and miss a few, one could say “so what.” I’m still probably getting more accuracy that if I was relying on a PEO entering data on their hand held. But what if I’m relying on LPR to allow vehicles to enter and exit a garage? If I miss 10% and don’t have any other way to get them in or out, I have a real problem in the lanes.
Granted the read rate may be higher in garages because the cars are stopped, the cameras aligned perfectly, and lighting is ‘just right.’ But still, some will be missed.
The point is that when you ask about read rates, be sure you ask the right questions. The answers may all be true, but not all are helpful.
JVH
3 Responses
Spot on John.
Nice synopsis.
Regards,
Bill
Just to be pedantic, UK number plates haven’t been white/silver on black since Adam was a child. Front plates are black characters on a retro reflective white background, rear are black characters on a yellow retro reflective background. This design is pretty similar across many European countries now; pity we left Europe.
Both front and rear are mandatory on everything except motor cycles which omits the front plate However, the point you make is very valid no one guarantees 100% scans even here and people that I trust tend to get nervous when you start to talk above about 97%. The thing is if you get 97% going in and 97% going out, your matching could be as low as 94%, so not yet perfect.
I remember may years ago talking to a wise man who identified a rather different problem; false reads. I understand that all number plate reading is done using probability. This collection of pixels represents an A, B, C. etc. with a certain probability if that is above the program threshold then it is the assessed character. Only it is of course quite possible to read a mud splat as a licence plate to the required level of “accuracy”. In a car park system this would be an inconvenience, when the camera is being used for law enforcement, the context of our conversation, with the threat of legal sanction, its much more serious. How do you guys out tere deal with this today?
Hi Peter –
Glad to have the record correct — The issue is that they are all the same. That makes reading much easier as the algorithm can be taught to work in all cases.
As for the issue of “misreads” I yield to those more expert than I.
JVH