2026-07-23T02:52:53Z https://drops.dagstuhl.de/oai

oai:drops-oai.dagstuhl.de:24460 2026-01-21T10:06:33Z ddc:004 open_access

QualiNet: Acquiring Bird’s Eye View Qualitative Spatial Representation from 2D Images in Automated Vehicle Perception (Short Paper) Belmecheri, Nassim Qualitative Spatial Representation Deep Learning Computer vision Qualitative Scene Understanding Spatio-temporal representation and reasoning models (including moving objects tracking) We present QualiNet, an end-to-end deep learning framework that acquires Bird’s Eye View (BEV) qualitative spatial relations directly from 2D images, eliminating the need for depth sensors. The system combines 2D object detection, masking, and classification to infer Rectangle Algebra (RA) and Qualitative Distance Calculus (QDC) relations. Evaluated on NuScenes and PandaSet datasets, QualiNet achieves 91% accuracy for RA, 80% for QDC, and 99% top-2 accuracy, demonstrating robust performance for automated vehicle perception. Schloss Dagstuhl – Leibniz-Zentrum für Informatik Nassim Belmecheri 2025 Is Part Of LIPIcs, Volume 355, 32nd International Symposium on Temporal Representation and Reasoning (TIME 2025) InProceedings Text doc-type:ResearchArticle publishedVersion application/pdf doi:10.4230/LIPIcs.TIME.2025.14 urn:nbn:de:0030-drops-244608 https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.TIME.2025.14 eng https://creativecommons.org/licenses/by/4.0/legalcode