\(\pi \) Match: Monocular vSLAM and Piecewise Planar Reconstruction Using Fast Plane Correspondences
This paper proposes \(\pi \) Match, a monocular SLAM pipeline that, in contrast to current state-of-the-art feature-based methods, provides a dense Piecewise Planar Reconstruction (PPR) of the scene. It builds on recent advances in planar segmentation fro
- PDF / 7,388,749 Bytes
- 16 Pages / 439.37 x 666.142 pts Page_size
- 4 Downloads / 154 Views
Abstract. This paper proposes πMatch, a monocular SLAM pipeline that, in contrast to current state-of-the-art feature-based methods, provides a dense Piecewise Planar Reconstruction (PPR) of the scene. It builds on recent advances in planar segmentation from affine correspondences (ACs) for generating motion hypotheses that are fed to a PEaRL framework which merges close motions and decides about multiple motion situations. Among the selected motions, the camera motion is identified and refined, allowing the subsequent refinement of the initial plane estimates. The high accuracy of this two-view approach allows a good scale estimation and a small drift in scale is observed, when compared to prior monocular methods. The final discrete optimization step provides an improved PPR of the scene. Experiments on the KITTI dataset show the accuracy of πMatch and that it robustly handles situations of multiple motions and pure rotation of the camera. A Matlab implementation of the pipeline runs in about 0.7 s per frame.
Keywords: Monocular visual SLAM
1
· Piecewise planar reconstruction
Introduction
Monocular Visual Simultaneous Localization and Mapping (vSLAM) is the process of estimating the camera position and orientation while building 3D maps of the environment, from a single camera. Although there has been intensive research on this topic, current methods still face several challenges and difficulties, including (i) presence of outliers, (ii) dynamic foregrounds and pure rotation of the camera, (iii) large baselines, (iv) scale drift, (v) density of 3D reconstruction, and (vi) computational efficiency. Nowadays, existing methods for monocular vSLAM follow two distinct approaches: feature extraction and direct image alignment. Each paradigm is effective in solving some of these challenges but, to the best of our knowledge, there is no monocular vSLAM algorithm that is able to tackle all these issues. While feature-based methods work on top of extracted features and are usually robust to outliers by applying RANSAC-based schemes [11], direct methods perform whole image alignment c Springer International Publishing AG 2016 B. Leibe et al. (Eds.): ECCV 2016, Part VIII, LNCS 9912, pp. 380–395, 2016. DOI: 10.1007/978-3-319-46484-8 23
πMatch: Monocular vSLAM and PPR Using Fast Plane Correspondences
(a) Img. pair w/ matched ACs
381
(b) PEaRL (0.22 s)
(c) T-Linkage (7.67 s)
(d) PEaRL (4.36 s)
(e) Affinity Prop. (0.20 s)
Fig. 1. Plane segmentation problem solved using different methods: (b) PEaRL [14] with 300 homography hypotheses, (c) T-linkage [18] with 300 hypotheses,(d) PEaRL with 5000 hypotheses, and (e) affinity propagation [7]. The computational times of PEaRL and T-linkage hamper real-time performance. On the contrary, affinity propagation is fast and is able to detect all the planes present in the image. Red points correspond to outliers. (Color figure online)
and cannot handle outliers [2,4,19]. Moreover, the former work with wide baselines and provide sparse reconstructions, as opposed to the latter that require small ba
Data Loading...
