RT info:eu-repo/semantics/conferenceObject
T1 Can we PASS beyond the Field of View? Panoramic Annular Semantic Segmentation for real-world surrounding perception
A1 Yang, Kailun
A1 Hu, Xinxin
A1 Bergasa Pascual, Luis Miguel
A1 Romera Carmena, Eduardo
A1 Huang, Xiao
A1 Sun, Dongming
A1 Wang, Kaiwei
K1 Electrónica
K1 Electronics
AB Pixel-wise semantic segmentation unifies distinctscene perception tasks in a coherent way, and has catalyzednotable progress in autonomous and assisted navigation, wherea whole surrounding perception is vital. However, current mainstream semantic segmenters are normally benchmarked againstdatasets with narrow Field of View (FoV), and most visionbased navigation systems use only a forward-view camera.In this paper, we propose a Panoramic Annular SemanticSegmentation (PASS) framework to perceive the entire surrounding based on a compact panoramic annular lens systemand an online panorama unfolding process. To facilitate thetraining of PASS models, we leverage conventional FoV imagingdatasets, bypassing the effort entailed to create dense panoramicannotations. To consistently exploit the rich contextual cues inthe unfolded panorama, we adapt our real-time ERF-PSPNet topredict semantically meaningful feature maps in different segments and fuse them to fulfill smooth and seamless panoramicscene parsing. Beyond the enlarged FoV, we extend focallength-related and style transfer-based data augmentations, torobustify the semantic segmenter against distortions and blursin panoramic imagery. A comprehensive variety of experimentsdemonstrates the qualified robustness of our proposal for realworld surrounding understanding.
PB IEEE
SN 2642-7214
YR 2019
FD 2019-08
LK http://hdl.handle.net/10017/45410
UL http://hdl.handle.net/10017/45410
LA eng
NO 2019 IEEE Intelligent Vehicles Symposium (IV), Paris, France, 9-12 June 2019
NO Ministerio de Economía y Competitividad
DS MINDS@UW
RD 29-abr-2024