Functions and definitions
OCR is the acronyms of “Optical Characters Recognizer” but with our Autovelox it would be more appropriate to call it ANRP (Automatic Number Plate Recognition) since it dose not simply recognize the characters but it recognizes and reads license plates.
The applications of the speed measuring device Autovelox 105SE are:
- reduce back office processing time by eliminating the need of manually inserting the license plate numbers
- automatically detect stolen cars by comparison of the internal database
Two kind of recognition methods can be selected during production:
In the “On Line” mode, during the normal operating of Autovelox 105SE, it automatically (and in real time) reads, after the image has been captured, each single license plate right. If duly configured for function n. 2 when the Autovelox archive is uploaded with information about “wanted and/or stolen cars”, the OCR, whenever one of those cars is found, will display a visible message on the screen together with an acoustic alarm.
The “Stolen Cars” database can be uploaded in the Autovelox by inserting in the CD drive of the Autovelox a CD previously prepared by an external software, the upload must be done during the Autovelox startup.
In the “Off Line” mode the infraction detection remains unchanged, while the plate recognition procedure will be activated only upon user’s request. This can be done at the end of the service when all images are stored on the Autovelox 105SE, before they are downloaded on the CD.
For function n. 1 license plate reading can be don easily both “Off Line” or “On Line”. On the “Off Line” mode the infraction detection is unchanged while the “On line” mode slightly reduces the Autovelox performances. In fact the OCR recognition time, which varies from 0.2 to 1 second (average 0.3 seconds), lengthens the idle time of the unit between the last detected infraction and the new “ready” state.
Which results should be expected
The character recognition done by a Computer follows logics which are totally different from the one followed by human. For this reason it is often hard to understand why the system did not recognize a plate that to us seems very clear. The same can happened in the opposite way, licenses that to us seem very unclear will be read easily by the OCR.
Said this we can give some general indications of factors which can negatively influence the OCR results:
- The license plate inclination
- When used from fixed site, we must make sure that the glass is clean and with no scratches hit by direct sun light
- Sun light reflection on the plate must be avoided
We must also consider those that are structural limits of the OCR which depend on its “Training”:
It recognizes only plates of a specific country at the time (production setup currently allows Italy and Spain)
It dose not recognize “square” plates
It dose not recognize plates with white numbers on black background
In order to obtain objective results with OCR we have generated a sample of 1600 images taken in various situations (day, night, etc…). on this sample we have obtained the following results:
True Positives: 90%
False Positives: 1,5%
The presence of one or more negative aspects (previously mentioned) determines a decrease in performance which sometimes can be significant and in any case unpredictable. For this reason the data contained in this document must be considered only as an indication of possible performance of the product OCR.