{"title":"Self-Supervised 3D Semantic Occupancy Prediction from Multi-View 2D Surround Images","authors":"S. Abualhanud, E. Erahan, M. Mehltretter","doi":"10.1007/s41064-024-00308-9","DOIUrl":null,"url":null,"abstract":"<p>An accurate 3D representation of the geometry and semantics of an environment builds the basis for a large variety of downstream tasks and is essential for autonomous driving related tasks such as path planning and obstacle avoidance. The focus of this work is put on 3D semantic occupancy prediction, i.e., the reconstruction of a scene as a voxel grid where each voxel is assigned both an occupancy and a semantic label. We present a Convolutional Neural Network-based method that utilizes multiple color images from a surround-view setup with minimal overlap, together with the associated interior and exterior camera parameters as input, to reconstruct the observed environment as a 3D semantic occupancy map. To account for the ill-posed nature of reconstructing a 3D representation from monocular 2D images, the image information is integrated over time: Under the assumption that the camera setup is moving, images from consecutive time steps are used to form a multi-view stereo setup. In exhaustive experiments, we investigate the challenges presented by dynamic objects and the possibilities of training the proposed method with either 3D or 2D reference data. Latter being motivated by the comparably higher costs of generating and annotating 3D ground truth data. Moreover, we present and investigate a novel self-supervised training scheme that does not require any geometric reference data, but only relies on sparse semantic ground truth. An evaluation on the Occ3D dataset, including a comparison against current state-of-the-art self-supervised methods from the literature, demonstrates the potential of our self-supervised variant.</p>","PeriodicalId":56035,"journal":{"name":"PFG-Journal of Photogrammetry Remote Sensing and Geoinformation Science","volume":"6 1","pages":""},"PeriodicalIF":2.1000,"publicationDate":"2024-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"PFG-Journal of Photogrammetry Remote Sensing and Geoinformation Science","FirstCategoryId":"89","ListUrlMain":"https://doi.org/10.1007/s41064-024-00308-9","RegionNum":4,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q3","JCRName":"IMAGING SCIENCE & PHOTOGRAPHIC TECHNOLOGY","Score":null,"Total":0}
引用次数: 0
Abstract
An accurate 3D representation of the geometry and semantics of an environment builds the basis for a large variety of downstream tasks and is essential for autonomous driving related tasks such as path planning and obstacle avoidance. The focus of this work is put on 3D semantic occupancy prediction, i.e., the reconstruction of a scene as a voxel grid where each voxel is assigned both an occupancy and a semantic label. We present a Convolutional Neural Network-based method that utilizes multiple color images from a surround-view setup with minimal overlap, together with the associated interior and exterior camera parameters as input, to reconstruct the observed environment as a 3D semantic occupancy map. To account for the ill-posed nature of reconstructing a 3D representation from monocular 2D images, the image information is integrated over time: Under the assumption that the camera setup is moving, images from consecutive time steps are used to form a multi-view stereo setup. In exhaustive experiments, we investigate the challenges presented by dynamic objects and the possibilities of training the proposed method with either 3D or 2D reference data. Latter being motivated by the comparably higher costs of generating and annotating 3D ground truth data. Moreover, we present and investigate a novel self-supervised training scheme that does not require any geometric reference data, but only relies on sparse semantic ground truth. An evaluation on the Occ3D dataset, including a comparison against current state-of-the-art self-supervised methods from the literature, demonstrates the potential of our self-supervised variant.
期刊介绍:
PFG is an international scholarly journal covering the progress and application of photogrammetric methods, remote sensing technology and the interconnected field of geoinformation science. It places special editorial emphasis on the communication of new methodologies in data acquisition and new approaches to optimized processing and interpretation of all types of data which were acquired by photogrammetric methods, remote sensing, image processing and the computer-aided interpretation of such data in general. The journal hence addresses both researchers and students of these disciplines at academic institutions and universities as well as the downstream users in both the private sector and public administration.
Founded in 1926 under the former name Bildmessung und Luftbildwesen, PFG is worldwide the oldest journal on photogrammetry. It is the official journal of the German Society for Photogrammetry, Remote Sensing and Geoinformation (DGPF).