2021/09/28 by Ali Safa, Tim Verbelen, Safa, Ali +9 · 1 citation
Computer Science · Engineering · #Advanced Neural Network Applications #Advanced SAR Imaging Techniques #Adversarial Robustness in Machine Learning #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Robotics (cs.RO)
paper · pdf · doi:10.48550/arxiv.2109.13666
openalex publication_date 2021/09/28 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Drones are currently being explored for safety-critical applications where\nhuman agents are expected to evolve in their vicinity. In such applications,\nrobust people avoidance must be provided by fusing a number of sensing\nmodalities in order to avoid collisions. Currently however, people detection\nsystems used on drones are solely based on standard cameras besides an emerging\nnumber of works discussing the fusion of imaging and event-based cameras. On\nthe other hand, radar-based systems provide up-most robustness towards\nenvironmental conditions but do not provide complete information on their own\nand have mainly been investigated in automotive contexts, not for drones. In\norder to enable the fusion of radars with both event-based and standard\ncameras, we present KUL-UAVSAFE, a first-of-its-kind dataset for the study of\nsafety-critical people detection by drones. In addition, we propose a baseline\nCNN architecture with cross-fusion highways and introduce a curriculum learning\nstrategy for multi-modal data termed SAUL, which greatly enhances the\nrobustness of the system towards hard RGB failures and provides a significant\ngain of 15% in peak F1 score compared to the use of BlackIn, previously\nproposed for cross-fusion networks. We demonstrate the real-time performance\nand feasibility of the approach by implementing the system in an edge-computing\nunit. We release our dataset and additional material in the project home page.\n