When building an automatized system, no matter for what purpose, the first step according to my approach is optimizing the quality of the input, and trying to minimize the number of challenging situations (or events) as much as possible. The equation is easy: optimum input quality for optimal performance; but what are the implications in case of LPR? The answer depends on the intended purpose of your system. What is the critical success factor of the system? Is it hitrate, maximum accuracy, or maximum processing speed? In our current series of entries I will be discussing some basic misunderstandings of LPR usually leading to lot more common misconceptions. Follow us to know what you can expect of your ALPR system to avoid post-implementation disappointment.
Misconception 1: In general, hardware based detection is more accurate than software based detection. – Wrong.
A precise software based detection doesn’t need to calculate with delay between the signal and the appearance of the vehicle in the monitoring area, hence synchronization problems don’t exist.
Misconception 2: Processing every single frame of a camera stream will provide the most accurate LPR results. – Wrong.
As a vehicle travels through the FOV (field of view) of the camera, the license plate appears in several frames. In the beginning and at the end of the session, it is often partially visible, and its size may vary a lot. How many different license plate results will be returned by the LPR engine?
Misconception 3: LPR systems read one image per vehicle, and create a database entry based on that sole reading – Wrong.
When building an automatized system, you have to consider including some redundancy. One of these is reading multiple images per vehicle and validating the LPR result.
Misconception 4: An LPR system should be able to detect fake license plates. – Wrong.
As if…License plate reading algorithms use an image of the license plate to read. Sometimes, fake license plates are very hard to detect based on an image even for a human. What current LPR systems can offer at most, is providing results with a “suspicion flag” in case any inconsistency is determined during the recognition.
Misconception 5: The LPR library is very resource-hungry, so decreasing its runtime must be the key factor for having a fast, very responsive system. – Wrong.
The LPR library has a very specific task which is the key feature of the complete system. This task needs to be done with special care, by taking the time that is necessary. I seriously advise not to tinker with the key processes when trying to cut back on resource demand.
To be continued…
English 
Deutsch