“Thanks to NVIDIA’s Deepstream SDK and TLT, we cut our initial development time significantly, allowing us to dedicate our technical resources to maximize the AI model compression rate and develop a real-time traffic signal control solution while improving its performance. As a result, Nota was able to further differentiate our solution efficiently and effectively to meet customers’ growing demands in time.”
Myungsu Chae, CEO of Nota

Nota Real-time Traffic Control Solution

Nota developed a real-time traffic control solution at Pyeongtaek, South Korea that had a real impact on drivers’ daily lives and saved social costs caused by traffic congestion. The On-device Intelligent Traffic Signal Control Technology developed by Nota uses image recognition technology to identify the traffic volume and queues, shares data to analyze the degree of congestion, and provides adequate traffic signal controls at the intersections. Most importantly, this solution runs seamlessly on the edge device thanks to its efficient AI model compression technology.NVIDIA PlatformNota is making roadways safer and more efficient by integrating NVIDIA’s edge GPUs and deep learning SDKs with its own in-house technology. For instance, Nota developed its intelligent video analytics application with NVIDIA’s acceleration toolkits—DeepStream and Transfer Learning Toolkit. DeepStream’s off-the-shelf features, such as line crossing and setting a region of interest, significantly improved the accuracy to examine the traffic situation and shortened development time. It also offers not only an optimized pipeline for video decoding and batched inference, but also tools for intelligent video analytics. DeepStream proves its versatile compatibility by supporting various video codecs, such as H.264, H.265, and MPEG-4. Easy development of pre-trained models led to better performance than training from scratch. The application was deployed on the Jetson AGX Xavier at a busy corridor to accurately count and detect the type of cars, understand traffic patterns, and provide insights to reduce congestion with adaptive traffic signal control systems. NVIDIA AGX Xavier is a well-suited candidate for IoT applications seeking a device at the best price with the highest and most reliable GPU-powered capabilities.Nota ResultsNota cut its development effort by 50% using NVIDIA software stackand building AI skills in three months rather than six months.Nota was able to reduce the model’s size down to 1/10 of its original size while maintaining high performance with mAP of 94.34. By analyzingtraffic flow and controlling traffic lights in real time, Nota was able tospeed up the average speed of cars by more than 300% during rush hour and by 25% during normal hours. Furthermore, adopting the AGX Xavier allowed Nota to reduce its hardware cost by 85% compared to the server-based solution that’s offered by most incumbents in the Intelligent Transportation System industry.

About Nota AI

Nota started at KAIST in 2015, with the goal of “making AI available everywhere through model compression.” NetsPresso was developed internally to automate AI model compression so vision-based AI models can be run on diverse hardware including a CPU, GPU, NPU, etc. The company leverages NetsPresso to provide the market with lightweight AI models to reduce latency, optimize processor usage, consume less energy, and proliferate lower-end edge devices while conserving accuracy. Nota has successfully launched its solutions for the Intelligent Transportation System and security industries. We’re active members of NVIDIA Inception and Metropolis, and are looking for more global partners to expand the market together.