Plus for plate number identification systems using Tapway AI and Nvidia GPU

Last November, Highway Concession Company PLUS Malaysia already announced that it had activated the Automatic Vehicle Registration Number Identification (ANPR) system to detect every vehicle entering the North-South Expressway through a toll plaza.

The system is enabled to ensure that each user provides a toll for highway use, and this is the first step before implementing a wireless toll with a multi-lane free flow (MLFF) system.

How does this automatic detection system work? Technology giant Nvidia has already released details of the ANPR Plus system, which uses computer vision and artificial intelligence (AI) technology, integrated with a graphics processing unit (GPU) and related software.

The company has created a system called VehicleTrack Startup Tapway from Malaysia. It is capable of reading vehicle registration plates as well as identifying vehicle class, brand and even color in just 50 milliseconds, even if the vehicle is driven at a speed of 40 km / h.

Tapway was founded by former astronaut Lim Chi Howe, and the company became involved in the project in 2019 after assisting in the video analysis process. Their client, PLUS, was looking for an effective way to track the entry and exit points of a vehicle for the North-South Expressway. Works with a closed system for tolls.

In particular, PLUS seeks to prevent customers from using different payment methods when entering and exiting highways – for example, using Radio Frequency Identification (RFID) but using Touch ‘n Go – when setting up payment systems using RFID technology. .

This is to prevent highway users from being charged twice or even trying to deceive the system. “We show them with computer vision – with just a camera and AI – you can do it all,” Lim said.

Tapway trains and manages their AI technology using NVIDIA A100 and V100 tensor core GPUs; As a result, the system is able to operate in all light and weather with consistent accuracy at 97%, Nvidia explained.

Each GPU is capable of handling 50 video streams at once using the Nvidia Triton interference server, which allows the ANPR system to process 28,800 images per minute per server using the Nvidia A10, A30 and T4 GPU.

“Triton is a real life saver for us,” Lim said. “We encounter scaling problems when guessing and Multithreading By itself and the server cannot exceed 12 video streams, but with Triton we are able to easily manage 20 and we tested 50 streams together, “he added.

According to Lim, Tapway uses 100,000 image data sets to deliver new AI models to customers in just a few hours, compared to a few days if a CPU-based system is used. The company built their applications using the Nvidia Dipstream Software Development Kit (SDK) and optimized their AI models with an SDK Tensor RT for high performance adaptive estimates.

As we mentioned earlier, PLUS uses the ANPR system not only in RFID lanes but also in Touch ‘n Go and SmartTAG lanes. This allows them to track all highway users, even if they enter using a TnG or SmartTAG card lane and exit with an RFID lane, thus charging eWallet (via RFID) and not from the TnG card. The company has so far installed a total of 577 cameras and plans to expand its use to 900 more in 92 toll plazas, Nvidia said.

But with all the technology, the ANPR PLUS system is still not ‘proof of failure’. Therefore, PLUS continues to provide a specialized verification center operated by their staff to help the system identify damaged, dirty or non-standard plate numbers using the alphabet, numbers and order. Fancy.

Since its inception in 2014, Tapwy has installed 3,000 sensors at 500 locations in Malaysia and Singapore. Their AI technology not only recognizes the registration number but also helps the mall conduct research on consumers’ buying habits. According to Invidia, the tapway seeks to assist Malaysian vehicle and palm oil manufacturers to improve quality control inspections.

Leave a Reply

Your email address will not be published.