Roadside radar classification improved using point cloud pattern analysis
Refining Roadside 4D Radar Vehicle Classification Using Point-Cloud Dispersion
Computer Vision and Pattern Recognition
Summary
Roadside sensors help track vehicles for smarter traffic systems, but usual cameras and older sensors don't work well in bad weather or at a distance. The authors show that 4D radar can see vehicles in all directions even in tough conditions, but it often confuses similar vehicles like cars and motorcycles because its data is noisy. They built a system combining radar and a stereo camera to gather real-world data and then found a clever way to look at how radar points spread over time. This method cuts errors in telling cars from motorcycles almost by a fifth and can work on devices with limited computing power.
What this means in practice
- •For traffic system operators: Improve roadside radar vehicle classification accuracy for safer and more reliable traffic monitoring under adverse conditions.
- •For autonomous vehicle developers: Use improved radar-only classification techniques to augment vehicle detection and classification in all weather environments.
Authors
Mohammed Alsakabi, Andreia Figueiredo, António Almeida, John Dolan, Miguel Luís, Pedro Rito, Ozan Tonguz, Susana Sargento
Abstract
Roadside sensing infrastructure is essential for Intelligent Transportation Systems (ITS), but cameras and legacy sensors are limited in range, weather robustness, and depth accuracy. 4D radar complements these modalities by resolving range, azimuth, elevation, and radial velocity under adverse conditions, though its sparse, noisy point clouds make reliable classification difficult. We present a roadside multi-sensor platform combining a 4D radar and stereo depth camera under synchronized acquisition, used to collect a real-world dataset. Our analysis reveals a substantial class-distribution discrepancy between modalities, most notably a strong radar confusion between passenger cars and motorcycles (49.01% and\ 48.60% of the dataset). To address this, we propose a Neyman-Pearson-based binary classifier on the temporally-smoothed 2D positional standard deviation of radar point cloud clusters, reducing classification error from 52.1% to 42.7%. This lightweight and edge-deployable approach improves radar-only classification reliability and lays the foundation for future camera-radar fusion.