A Dense Segmentation Network for Fine Semantic Mapping
Authors:
- 1 December 2019
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1969-1974
Abstract
This paper proposes a fine semantic mapping method using dense segmentation network (DS-Net) to obtain good performance of semantic mapping fusion, in which the semantic segmentation network (DS-Net) is constructed based on the idea of DenseNet’s dense connection. First, the RGB image and the depth image are used to generate a dense indoor scene map via the state-of-the-art dense SLAM (ElasticFusion). Then, semantic segmentation are precisely performed on the input RGB image via DS-Net. Finally, the long-term correspondence between the landmarks and the indoor scene map is established using the continuous frames both in the visual odometer and loop detection, and the final fused semantic map is obtained by integrating semantic predictions of the RGB-D video frames of multiple angles with the indoor scene map. Experiments were performed on the NYUv2 and CIFAR10 datasets and our laboratory environments. Results show shows that our method can reduce the error of dense map construction and obtain good semantic segmentation performance.Keywords
This publication has 13 references indexed in Scilit:
- BundleFusionACM Transactions on Graphics, 2017
- SemanticFusion: Dense 3D semantic mapping with convolutional neural networks2017
- Probabilistic data association for semantic SLAM2017
- Past, Present, and Future of Simultaneous Localization and Mapping: Toward the Robust-Perception AgeIEEE Transactions on Robotics, 2016
- ElasticFusion: Real-time dense SLAM and light source estimationThe International Journal of Robotics Research, 2016
- Learning Deconvolution Network for Semantic Segmentation2015
- Incremental and batch planar simplification of dense point cloud mapsRobotics and Autonomous Systems, 2015
- Fully convolutional networks for semantic segmentation2015
- Dense 3D semantic mapping of indoor scenes from RGB-D images2014
- RGB-D mapping: Using Kinect-style depth cameras for dense 3D modeling of indoor environmentsThe International Journal of Robotics Research, 2012