Working as an aerospace engineer in Malaysia, Chee How Lim dreamed of creating a startup that could really take off. Today, his company, Tapway, is riding the wave of computer vision and AI adoption in Southeast Asia.
A call for help in 2019 with video analytics led to the Kuala Lumpur-based company’s biggest project to date.
Malaysia’s largest toll road operator, PLUS, wanted to reduce congestion for its more than 1.5 million daily travellers. A national plan called for allowing the movement of cars, taxis, buses and trucks to move freely in multiple lanes, but this posed several significant challenges.
Relieve traffic jams
Motorways charge five classes of tolls depending on the type of vehicle. Drivers pay using four different systems and often enter the highway using one payment system and then exit using another, making it difficult to track vehicles.
Dedicated lanes for different categories of vehicles forced users to stop, slowing traffic, so pit operators could also identify the specific vehicle. Even then, some drivers ripped off the system, swapping cards on the highway to get lower tolls.
“We showed them how with computer vision — just a camera and artificial intelligence — you could solve all of this,” Lim said.
AI smooths the flow
Using GPUs and NVIDIA software, Tapway trained and ran AI models that could read a vehicle’s license plate and detect its class, make, and color in just about 50 milliseconds. a tenth of an eye blink, even if it travels up to 40 kilometers per hour when approaching a toll plaza.
Tapway’s VehicleTrack software works in all lighting and weather conditions with a consistent 97% accuracy. And thanks in part to NVIDIA Triton Inference Servera single GPU can handle up to 50 simultaneous video streams.
PLUS has installed 577 cameras to date and plans to extend nearly 900 of them to 92 toll booths to achieve its goal of smoothing traffic.
Inside a computer vision system
Under the hood, the system relies on intelligent AI models trained in the cloud on an array of NVIDIA A100 and V100 Tensor Core GPUs.
They use a dataset of up to 100,000 images to prepare a new model for a Tapway client in hours, a huge improvement over a CPU-based system that took days, Lim said.
But the real magic comes with the inference, running these models in production to process up to 28,800 frames per minute on edge servers using NVIDIA A10, A30, and T4 GPUs.
The software makes it sing
Faucet Uses NVIDIA DeepStream software development kit to build its computer vision applications, NVIDIA TensorRT to keep its AI models light and fast, and Triton to play the role of the traffic cop, directing the AI inference work.
“Triton is a real lifeline for us,” Lim said. “We had some scaling issues when inferencing and multithreading on our own and couldn’t scale beyond 12 video streams in a server, but with Triton we easily manage 20 and we tested it on up to 50 simultaneous streams,” he said.
In February, Tapway officially became a NVIDIA Metropolis partner. The program provides smart video analytics companies with rapid access to technology and expertise.
“We’ve had to pass stress tests in areas such as multistreaming and security – which have helped strengthen our product offering – and from a business perspective, it’s a way to gain recognition and to further establish ourselves as an AI expert in the region,” Lim said.
AI covers the waterfront
Since its inception in 2014, Tapway has deployed 3,000 sensors in 500 locations in Malaysia and Singapore. Off the road, they’re helping malls and retailers understand customer buying habits, and now the company is gearing up to help manufacturers like automakers and palm oil producers in the region inspect products for quality control.
“The demand has never been better, there are a lot of vision challenges in the world and quite a few exciting projects that we hope to see soon,” he said.