CNN Unrolled Sparse Bayesian Learning for Automotive Occupancy Grid Mapping
IEEE Sensors Journal · 2026-08 · 26 (15) · 23514-23522
DOI: 10.1109/jsen.2026.3703917 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Sensors Journal, 2026-08. Verfasst von Nikolaos Fotopoulos und 3 weiteren Autor:innen. Die Arbeit nennt 25 Literaturangaben. Gefördert von: Rijksdienst voor Ondernemend Nederland.
Abstract (Originalsprache)
Automotive occupancy grid mapping creates a discretized grid-based map of a vehicle’s environment using onboard sensor data, with each grid cell labeled as occupied or free of obstacles. In this work, we address the problem of estimating an occupancy grid map from radar and LiDAR point cloud data. Traditional approaches, such as pattern-coupled sparse Bayesian learning (PC-SBL), exploit the block-sparse, spatially correlated structure of the maps via explicit priors that are not data-adaptive and incur high computational cost. We present a hybrid approach that combines physics-informed Bayesian methods with data-driven deep learning. Specifically, we incorporate a lightweight convolutional neural network as a learnable implicit prior in a deep unfolded PC-SBL method, allowing the model to adaptively learn spatial correlations. Qualitative and quantitative evaluations on the nuScenes and RADIATE public datasets show that our approach outperforms the baseline PC-SBL in both detection performance and runtime.
Förderung
- Rijksdienst voor Ondernemend Nederland
Zitieren
APA
Fotopoulos, N., Joseph, G., Onen, C., & Pandharipande, A. (2026). CNN Unrolled Sparse Bayesian Learning for Automotive Occupancy Grid Mapping. IEEE Sensors Journal, 26(15), 23514-23522. https://doi.org/10.1109/jsen.2026.3703917
Vancouver
Fotopoulos N, Joseph G, Onen C, Pandharipande A. CNN Unrolled Sparse Bayesian Learning for Automotive Occupancy Grid Mapping. IEEE Sensors Journal. 2026;26(15):23514-23522. doi:10.1109/jsen.2026.3703917
BibTeX
@article{Fotopoulos2026,
author = {Fotopoulos, Nikolaos and Joseph, Geethu and Onen, Cagan and Pandharipande, Ashish},
title = {CNN Unrolled Sparse Bayesian Learning for Automotive Occupancy Grid Mapping},
journal = {IEEE Sensors Journal},
year = {2026},
volume = {26},
number = {15},
pages = {23514--23522},
doi = {10.1109/jsen.2026.3703917}
}
EUVIMED