Multi-Radar Fusion for Failure-tolerant Vulnerable Road Users Classification

Abstract

Data fusion is one of the key aspects in robust and failure-tolerant vulnerable road user (VRU) perception systems. This paper presents a multi-radar sensor fusion platform that enables automatic detection, tracking and classification of pedestrians and cyclists while aiming to support fail-safe system operation. The sensor fusion platform encapsulates two main modules working concurrently: the first detection-to-detection fusion module performs a spatio-temporal alignment of radar detections, data association of the aligned detections and finally multi-object tracking; the second module is the sensor failure management block, which is responsible for fault-tolerant system operation and encapsulates multilevel verification subsystem. The proposed multi-radar fusion system is experimentally evaluated through multi-target scenarios. Demonstrated results show effectiveness of the proposed platform and failure-tolerance of the system compared to a single sensor solution.