Lidar specialist Ouster has announced that the deep learning perception model underlying its BlueCity traffic management system has reached a training milestone of 4 million labelled objects drawn from lidar data from 800 separate sites. These have provided BlueCity’s AI with training based on a wide range of traffic patterns, intersection designs, and environmental conditions

Ouster BlueCity employs proprietary perception software incorporating a deep neural network (DNN) to detect, classify, and track objects and trajectory data of multimodal road users. Ouster BlueCity’s DNN runs on NVIDIA Jetson AGX Orin and NVIDIA Orin NX system-on-modules for real-time inference on the edge, bringing physical AI to smart city traffic systems around the world. The company’s deep learning model is developed in-house and validated against manually annotated lidar data for accuracy.

Ouster BlueCity’s edge devices can process large amounts of 3D lidar data in real time. They deliver low-latency object detection, classification, and tracking, as well as supporting V2X communications and other intelligent transportation system (ITS) applications.

Compared to classical algorithms, this approach provides improved generalisation, greater scalability, reduced computational resources and real-time processing without the need for calibration.

“Ouster BlueCity is a prime example of how we are working to solve important real-world problems with physical AI,” said Ouster CEO Angus Pacala. “Powered by the NVIDIA Jetson Platform, Ouster BlueCity combines digital lidar with real-time perception to improve traffic management and road safety for all road users.

In 2024, Ouster closed deals to expand the adoption of Ouster BlueCity to over 400 sites globally, including the largest lidar-powered smart traffic network in the United States in Chattanooga, Tennessee.