In November, Highway Concession Plus Malaysia announced that it had introduced automatic number plate recognition (ANPR) to track vehicles passing through the North-South Highway Toll Plaza. This system is considered to be the first step towards ensuring that motorists are paying their toll fees as they should and this is considered a first step towards the goal of multi-lane free flow (MLFF), a form of open road tolling that is shared with common booths.
Now, technology giant Nvidia PLUS has detailed the computer vision and artificial intelligence (AI) technology found in the ANPR system, which includes the company’s proprietary graphics processing units (GPUs) and software. VehicleTrack, developed by Malaysian startup Tapway, can read a number plate and detect the class, making and color of a vehicle in just 50 milliseconds – even when it travels at a speed of 40 km / h.
Founded by former astronaut Lim Chi Howe, Tapway entered the project in 2019 in response to a call for help with video analysis. The client, PLUS, wanted to be able to effectively track the entry and exit points of a vehicle. The North-South Highway operates a closed system for tolls.
Specifically, the company sought to prevent users from using one payment method to enter the highway and another to exit, presumably in preparation for the rollout of Touch ‘n Go’s Radio Frequency Identification (RFID). This was done to prevent motorists from trying to cheat the system or double charging. “We showed them how with computer vision – just a camera and AI – you can solve all this,” Lim said.
Tapway trains and operates its AI models using NVIDIA A100 and V100 tensor core GPUs; The result is a system that works with a consistent 97% accuracy in all lighting and weather, Nvidia says. Each GPU can handle up to 50 video streams at once. Thanks to an Nvidia Triton Interference server, ANPR enables the system to process 28,800 images per minute on the edge server using Nvidia A10, A30 and T4 GPUs.
“Triton is a real life saver for us,” Lim said. “We had some scaling issues with guessing and multithreading ourselves and couldn’t scale more than 12 video streams on one server, but with Triton we could easily handle 20 and we tested it in 50 simultaneous streams,” he said.
According to Lim, Tapway uses up to 100,000 image datasets to create a new AI model for a customer in just a few hours instead of a few days for a CPU-based system. The company builds its app using the Nvidia Dipstream Software Development Kit (SDK) and optimizes its AI models through TensorRT, an SDK for high-performance deep learning estimates.
As we mentioned earlier, PLUS uses Touch ‘n Go and SmartTAG and ANPR systems, not just RFID lanes. It enables tracking of all users, even if they log in using a TnG card or SmartTAG and exit using RFID, charging their eWallet (via RFID) instead of their card. The company has so far installed 577 cameras and plans to increase that number to about 900 in 92 toll plazas, Nvidia reported.
Even with all these technologies, PLUS’s ANPR system is not stupid To that end, Concession has implemented a dedicated validation center where backend workers are able to further improve the system by identifying damaged, dirty or non-standard (such as “fancy” fonts and spacing) plates.
Since its inception in 2014, Tapway has implemented 3,000 sensors at 500 locations across Malaysia and Singapore. Its AI technology not only identifies number plates but also helps malls study consumer buying habits; The company even wants to help regional carmakers and palm oil producers improve quality control inspections, Nvidia said.