In recent years some direct monocular SLAM methods have appeared achieving impressive semi-dense or dense 3D scene reconstruction. At the same time, feature-based monocular SLAM methods can obtain more accurate trajectory than direct methods, but only obtain sparse feature point map rather than semi-dense or even dense map like direct methods. With the development of deep learning, it becomes possible to predict the depth map of a scene given a single RGB image. In this paper we demonstrate how depth prediction module via deep learning can be used as a plug-in module in highly accurate feature-based monocular SLAM (e.g. ORB-SLAM). Both accurate trajectory from ORB-SLAM and dense 3D reconstruction from depth prediction can be achieved. Evaluation results show that dense scene reconstruction can be obtained from highly accuarate feature-based monocular SLAM.
Dense Mapping from Feature-Based Monocular SLAM Based on Depth Prediction
01.08.2018
443552 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
ORB-SLAM based semi-dense mapping with monocular camera
IEEE | 2017
|Sensitivity study for feature-based monocular 3D SLAM
IEEE | 2015
|Vision-Based SLAM: Stereo and Monocular Approaches
British Library Online Contents | 2007
|CNN-Based Dense Monocular Visual SLAM for Real-Time UAV Exploration in Emergency Conditions
DOAJ | 2022
|