Pedestrian recognition with radar can enable the deployment of advanced driver-assistance systems. Despite neural networks representing a powerful yet effective tool for addressing this capability, their abilities as well as their limits have been scarcely
addressed in the literature.
In this paper, we investigate how point-wise processing architectures use radar features to perform the task of pedestrian detection. We study the behavior of four different techniques in reaction to a perturbation of radar point clouds. We show that
radar Doppler represents the most critical feature for the detection of pedestrians. However, we note that context plays an important role.
Finally, we prove that PointNet++ learns to use all the radar features in a proper, meaningful way, thus achieving superior performance to PointNet and Random Forest.
Authors: Alessandro Cennamo, Florian Kaestner, Anton Kummert
Publication Date: February 20, 2021
Published In: Association for Computing Machinery
Research Area: Machine learning, neural networks, environmental model
