Stockholm Toll System
The IBM Stockholm Road Charging system manages road toll collection for vehicles entering the inner city boundaries of Stockholm. The toll charges are based on vehicle license plates that are read when vehicles enter the city. Since manual interpretation of camera images is costly, the city of Stockholm requires very high automated recognition rates and very few false positives. When existing OCR solutions did not meet these criteria, researchers at the IBM Haifa Lab were approached to find a solution.
A pilot project reduced traffic by 25 percent, increased public transport usage by 40,000 users per day, and dramatically cut peak-time road congestion. This success was enabled by substantially reducing the recognition error rate. The success of the pilot project gave rise to a further deployment of this solution.
This project was mentioned recently in the article "IBM's grand plan to save the planet", which appeared on CNN in April 2009.
Research Work
The Haifa research team delivered an OCR engine that increased the number of license plates identified by approximately 20%, significantly reducing data entry costs.
The enhanced IBM solution starts by examining the photographs of the license plates and attempting to identify the car number. If the complete number is identified immediately, it is recorded in the system and stored for further business processing. If identification fails, the picture is moved to a central server where sophisticated algorithms make a second attempt at identification using techniques such as image enhancement, comparison of the front and back plates to make sure they correspond, and comparing results to those of a second OCR system.
By incorporating the Haifa technology to produce a system with two OCRs, the Stockholm team achieved the required accuracy that would have not been possible with a single OCR.
- The system has sophisticated recognition capabilities to help recognize license plate numbers, even at night or under bad weather conditions
- The engine analyzes the image and searches for predefined patterns and symbols, mimicking the human approach for deciphering images
- The system automatically records the registration number and processes the payment or billing
The result is a substantially reduced error rate.
The following diagram presents a flowchart of the toll system.
Figure 25 - System scheme
Example
Image received has low contrast
- Enhancements are applied
- License plate is located
- After proper binarization, the text is extracted via OCR
-> MA-323BA
Figure 26 - From top to bottom: a) Initial image b) Plate image c) Binarized plate and OCR recognition