Surviving Dominant Planes in Uncalibrated Structure and Motion Recovery

Abstract

In this paper we address the problem of uncalibrated structure and motion recovery from image sequences that contain dominant planes in some of the views. Traditional approaches fail when the features common to three consecutive views are all located on a plane. This happens because in the uncalibrated case there is a fundamental ambiguity in relating the structure before and after the plane. This is, however, a situation that is often hard to avoid in man-made environments. We propose a complete approach that detects the problem and defers the computation of parameters that are ambiguous in projective space (i.e. the registration between partial reconstructions only sharing a common plane and poses of cameras only seeing planar features) till after self-calibration. Also a new linear self-calibration algorithm is proposed that couples the intrinsics between multiple subsequences. The final result is a complete metric 3D reconstruction of both structure and motion for the whole sequence. Experimental results on real image sequences show that the approach yields very good results.

Globally Optimal Algorithms for Stratified Autocalibration

“…8(b)). The same angle is measured as 89.8 • in a reconstruction using (Pollefeys et al 2002). The precise ground truth angle between the targets is unknown.…”

“…9(b). The metric reconstruction using a locally optimal linear algorithm for directly estimating the absolute quadric (Pollefeys et al 2002) is shown in Fig. 9(c).…”

Calibrating Distributed Camera Networks

“…However, nodes that share only a few corresponding points with node i are removed from the cluster in order to ensure a minimum nucleus of corresponding points seen by all cameras in the cluster. The minimum cluster and nucleus size required for viable calibration depend on the number of parameters to be estimated at each node [67], [68]. In our experiments, we use a minimum cluster size of 3 cameras and a nucleus of at least 8 corresponding points.…”

“…The approach we favor is based on estimating a 4 × 4 symmetric rank-3 matrix called the absolute dual quadric, which is constrained by the camera matrices and some known properties of each camera's fixed internal parameters (for example, that the elements of the CCD array are square). Detailed descriptions of metricfrom-perspective recovery based on this approach are given by Pollefeys et al [67], [68].…”

Horizon constraint for unambiguous UAV navigation in planar scenes

“…The chirality constraints can then be applied to resolve this ambiguity unless all points are closer to one viewpoint than the other, in which the motion would be unresolvable [1]. Several methods such as [2], [6], [7] were proposed in order to overcome the planar degeneracy in the epipolar geometry-based SFM by incorporating the knowledge of the coplanarity in the algorithm design. Such works provide sufficient solution to the problem since scene regularities can generally be detected [8].…”