ALPR – It Takes More Than a $350 Camera
May, 2013
Editor, Parking Today:
I saw your piece in the March 2013 issue of Parking Today – “Technology, Friend or Foe?” – and thought you made a couple of very good points.
License plate recognition is limited, of course, because it is asked to score its own results based on a confidence level it applies to each “read’ event. We all (suppliers of the technology) use a suite of rules we’ve developed to help us with how we apply that confidence level during the ALPR process.
For example, if we know the license plates we are likely to see in a certain location always use a letter, as opposed to a number, in the space occupied by the third character on the license plate, and a vehicle from another state is encountered by the system that has the number “1” in that location on the plate, then it might be incorrectly read as the letter “I.”
This is an oversimplification of the problem, but sort of gets the point across as to the difficulty in providing 100% accuracy when reading license plates. This example is true for LPI systems as well as LPR systems, so I am uncertain how running both at an airport parking facility lets you know immediately how many misreads you will get, as you have suggested.
But let’s consider the gravity of the problem with the letter “I” and the number “1” example; there really shouldn’t be a problem at all in a parking facility. If the license plate is incorrectly read entering and exiting, it is still identified as the vehicle assigned to a particular ticket number or, if the system uses only license plate reads and not a ticket, it is still the same vehicle.
Many airport parking designs call for accuracy of a certain number of the characters on a license plate to be correct, say five out of seven. If the system has a ticket issued at the entry point, and five out of seven characters – in the same order – match on the license plate at the exit point, you can rest assured that you have the same vehicle associated with that ticket as the one to which it was issued on entry.
I know there are going to be people who say, “Oh yeah, what if a husband and wife swap tickets, and they registered their cars together, and the plates are only one digit apart?” Well, then, OK, you’ve got me.
But we all know that, statistically, that’s not going to happen. The thing to remember in a parking application is that you already know which vehicles have entered the facility when reading a license plate at the exit point – you’ve already seen them when they came in.
So it is not as difficult as perhaps on a toll road, where we are reading different plates all the time; we have to find other means of cleverness for those systems – and we do. In toll road applications, we have to guarantee that we identify the registered vehicle owner correctly; in parking, we are really guaranteeing the vehicle at the exit cashier is the one we say came in at a certain day and time, or belongs with a certain ticket number.
And thank you for your second suggestion for minimizing problems regarding expectations from a $350 video camera. Most people have no idea how difficult it can be to capture an image from a vehicle, under a variety of lighting conditions, that will present useable information to the OCR piece of an ALPR system.
The October 2004 issue of Parking Today published the first of a two-part series on these, and other points of ALPR in parking (“Applications of ALPR In Parking Facilities”). I think the information in that article was timely then and still is today. I will gladly forward a copy to anyone who would like to read it.
Thanks, and thanks for a great magazine.
Jim Kennedy
President/CEO, INEX Technologies
email: mk@inexzamir.com
I saw your piece in the March 2013 issue of Parking Today – “Technology, Friend or Foe?” – and thought you made a couple of very good points.
License plate recognition is limited, of course, because it is asked to score its own results based on a confidence level it applies to each “read’ event. We all (suppliers of the technology) use a suite of rules we’ve developed to help us with how we apply that confidence level during the ALPR process.
For example, if we know the license plates we are likely to see in a certain location always use a letter, as opposed to a number, in the space occupied by the third character on the license plate, and a vehicle from another state is encountered by the system that has the number “1” in that location on the plate, then it might be incorrectly read as the letter “I.”
This is an oversimplification of the problem, but sort of gets the point across as to the difficulty in providing 100% accuracy when reading license plates. This example is true for LPI systems as well as LPR systems, so I am uncertain how running both at an airport parking facility lets you know immediately how many misreads you will get, as you have suggested.
But let’s consider the gravity of the problem with the letter “I” and the number “1” example; there really shouldn’t be a problem at all in a parking facility. If the license plate is incorrectly read entering and exiting, it is still identified as the vehicle assigned to a particular ticket number or, if the system uses only license plate reads and not a ticket, it is still the same vehicle.
Many airport parking designs call for accuracy of a certain number of the characters on a license plate to be correct, say five out of seven. If the system has a ticket issued at the entry point, and five out of seven characters – in the same order – match on the license plate at the exit point, you can rest assured that you have the same vehicle associated with that ticket as the one to which it was issued on entry.
I know there are going to be people who say, “Oh yeah, what if a husband and wife swap tickets, and they registered their cars together, and the plates are only one digit apart?” Well, then, OK, you’ve got me.
But we all know that, statistically, that’s not going to happen. The thing to remember in a parking application is that you already know which vehicles have entered the facility when reading a license plate at the exit point – you’ve already seen them when they came in.
So it is not as difficult as perhaps on a toll road, where we are reading different plates all the time; we have to find other means of cleverness for those systems – and we do. In toll road applications, we have to guarantee that we identify the registered vehicle owner correctly; in parking, we are really guaranteeing the vehicle at the exit cashier is the one we say came in at a certain day and time, or belongs with a certain ticket number.
And thank you for your second suggestion for minimizing problems regarding expectations from a $350 video camera. Most people have no idea how difficult it can be to capture an image from a vehicle, under a variety of lighting conditions, that will present useable information to the OCR piece of an ALPR system.
The October 2004 issue of Parking Today published the first of a two-part series on these, and other points of ALPR in parking (“Applications of ALPR In Parking Facilities”). I think the information in that article was timely then and still is today. I will gladly forward a copy to anyone who would like to read it.
Thanks, and thanks for a great magazine.
Jim Kennedy
President/CEO, INEX Technologies
email: mk@inexzamir.com
