Degeneracies in Rolling Shutter SfM
We address the problem of Structure from Motion (SfM) with rolling shutter cameras. We first show that many common camera configurations, e.g. cameras with parallel readout directions, become critical and allow for a large class of ambiguities in multi-vi
- PDF / 4,468,836 Bytes
- 16 Pages / 439.37 x 666.142 pts Page_size
- 71 Downloads / 193 Views
Czech Technical University in Prague, Prague, Czech Republic {alblcene,pajdla}@cmp.felk.cvut.cz 2 National Institute of Informatics, Tokyo, Japan [email protected]
Abstract. We address the problem of Structure from Motion (SfM) with rolling shutter cameras. We first show that many common camera configurations, e.g. cameras with parallel readout directions, become critical and allow for a large class of ambiguities in multi-view reconstruction. We provide mathematical analysis for one, two and some multiview cases and verify it by synthetic experiments. Next, we demonstrate that bundle adjustment with rolling shutter cameras, which are close to critical configurations, may still produce drastically deformed reconstructions. Finally, we provide practical recipes how to photograph with rolling shutter cameras to avoid scene deformations in SfM. We evaluate the recipes and provide a quantitative analysis of their performance in real experiments. Our results show how to reconstruct correct 3D models with rolling shutter cameras. Keywords: Structure from motion Non-perspective cameras
1
·
Rolling shutter
·
Degeneracy
·
Introduction
Structure from Motion (SfM) reconstructs geometry of scenes from their images while simultaneously estimating camera poses and (some of) their internal parameters [5]. SfM has many practical applications in scene modelling, 3D mapping, and visual odometry [12,16,18]. Typical SfM considers perspective cameras, incrementally performs [16] and includes relative and absolute camera pose computation and bundle adjustment (BA) [17]. Recently, rolling shutter cameras became very important [11] since the rolling shutter is present in vast majority of current CMOS image sensors in consumer cameras and smart-phones. In rolling shutter cameras, images are not captured at once. They are scanned either along image rows or columns [13]. Since different image lines are exposed at different times, camera movement during the exposure produces image distortions. It has been shown that rolling shutter distortion can severely influence SfM computation [6,14] and that special care has to be taken to achieve sensible results. Authors of [6] addressed the problem of SfM from video sequences and presented specially adapted BA algorithm for rolling shutter videos [7]. In [9] a SfM pipeline for cellphone videos is presented using video sequences and fusion with c Springer International Publishing AG 2016 B. Leibe et al. (Eds.): ECCV 2016, Part V, LNCS 9909, pp. 36–51, 2016. DOI: 10.1007/978-3-319-46454-1 3
Degeneracies in Rolling Shutter SfM
37
Fig. 1. SfM with rolling shutter model can deliver undesired results. (Left) A reconstruction from forward camera translation with vertical readout direction. (Middle) A reconstruction of the same scene from forward moving camera horizontal readout direction. In both cases, the scene collapses into a plane that is perpendicular to the readout direction. (Right) When both image directions are combined a correct reconstruction is obtained with rolling shutter (RS) projec
Data Loading...
