Ekaterina Mutina, Marketing Manager for TRASSIR, looks at how automatic recognition systems for vehicles help keep businesses running smoothly and safely and support smart city management.

The use of automatic license plate recognition systems has become more widespread in the Middle East. This has been partially fueled by the growing trend of implementing smart cities and the advances in artificial intelligence technology which helps power these systems. 

In order to meet smart city objectives, it’s vital that the necessary elements for managing traffic conditions and increasing surveillance and safety on the roads to avoid incidents are in place.  Automatic number plate recognition (ANPR) systems have already been in use for decades, predominantly by police forces. Nowadays, many businesses already rely heavily on these security systems. After all, these technologies have already proven themselves and are successfully used for a range of tasks:

• Restricting access to the territory, as there are territories not everyone is permitted to enter, 
• Organizing parking, including paid parking lots in shopping and business centers, private and intercepting parking lots, and many others, 
• Organizing payment for parking services, where license plate recognition can automate the payment process, 
• Managing traffic flow, so you can flexibly configure access levels and create territories where only certain types of transport are allowed to enter, 
• Managing time vehicles spend on the territory at in-demand locations such as airports, train and metro stations, transport hubs and residential territories, 
• Registering vehicles to collect statistics that allow you to analyze traffic congestion, 
• Tracking vehicles, where auto-recognition systems can track the appearance of vehicles from watch lists created specially for this purpose and issue alarm signals when they appear.

Security capabilities

There are different versions of the vehicle recognition system available on the market, but as a rule, their hardware and software are based on Al technologies. Generally, these systems have a wide range of capabilities, such as automatically detecting and determining the vehicle type and license plate in real-time and then storing and cross referencing it to white/black lists.

Al for LPR

The degree to which Al technologies are applied plays a great role in the work of vehicle detection systems and many tasks are solved more efficiently with Al than with standard mechanisms of the past.

For example, the TRASSIR system's neural network is able to accurately determine the coordinates of a license plate's corners (even if it is located at an angle to the camera), which allows it to record the car's data as accurately as possible. Experience has proven that the system's implementation allows users to save money on the maintenance and development of facilities in the fields of construction, retail and industry. In some cases, vehicle control time at the checkpoint was reduced by half.

The latest improvements in the field of auto detection have been aimed at improving vehicle tracking: since vehicles can differ visually from one another greatly, the neural network calculates a vector of unique features of the car to improve recognition accuracy.

Market trends are such that systems are constantly being improved, as manufacturers continuously collect feedback from users. We can look at the following areas to improve the module:

• Training on new data, 
• Applying the latest neural network architectures, 
• Improving the quality monitoring process, 
• Other unique developments. 

The six steps of license plate recognition.

As a rule of thumb, most license plate recognition algorithms follow the steps below:

1. Firstly, the LPR engine looks to identify the license plate's positioning within the image - this helps the LPR focus only on the license plate and disregard any other data.

2. Angular corrections help the LRP decode license plates that have been captured at awkward angles - for example, on the side or from above.

3. Filters are then applied to help eliminate any shadows or shaded areas. Edge detection in particular is used when there is high contrast between the background and the text being identified.

4. Tools like whitespace delineation are used to identify the spaces between letters on the license plate. Errors are more likely to occur here if the spacing of characters on the license plate varies.

5. Optical character recognition techniques are then used to identify each character. This could include pattern matching, proportion, pixel repetition, and edge tracing.

6. In the final step the characters identified and their sequence are checked against rules specific to each region.

Follow the link below to learn more:

https://www.securitymiddleeastmag.com/magazine/