M2-2.1 - Enhancing Radar-Based Pedestrian Detection with Artificial Intelligence for Multipath Mitigation

Event

23. ITG/GMA-Fachtagung Sensoren und Messsysteme 2026
2026-06-09 - 2026-06-10
Nürnberg

Band

Vorträge

Chapter

Modellbildung und Signalverarbeitung

Author(s)

I. Hoyer, S. Böller, K. Seidl - Fraunhofer IMS, Duisburg

Pages

398 - 403

DOI

10.5162/sensoren2026/M2-2.1

ISBN

978-3-910600-11-9

Price

free

Abstract

Pedestrian safety in automated road traffic requires detection systems that function under complex environmental conditions. Camera-based systems may fail in fog, darkness, or occlusion, making radar an attractive alternative. This work investigates passive radar transponders integrated into pedestrian clothing or footwear, detected by a vehicle-mounted radar, to enable robust pedestrian localization. A key challenge is multipath propagation, which distorts the backscattered signal and degrades distance and angle estimation accuracy. To address this, an AI-based approach is proposed: a convolutional neural network (CNN) for distance estimation and a fully connected feedforward neural network (FCFNN) for angle estimation. Both models are trained and evaluated on synthetically generated radar datasets covering distances up to 10 m and angles up to 90°, with scenarios ranging from a single reflection up to 50 multipath reflections, including a realistic case with a random number of reflections and added noise at 23 dB SNR. The AI-based approach is benchmarked against conventional signal processing algorithms. In the scenario with random number of reflections, distance and noise, the proposed models achieve a mean absolute error (MAE) of 0.29 m for distance and 7.50° for angle estimation, compared to 0.40 m and 8.19° for the conventional method—improvements of approximately 28 % and 9 %, respectively. In most tested multipath conditions, the AI-based approach outperforms the conventional baseline. Future work will focus on real radar measurement validation, hyperparameter optimization, and extension to higher distances and moving objects.