Most man-made environments, such as urban and indoor scenes, consist of a setof parallel and orthogonal planar structures. These structures are approximatedby Manhattan world assumption and be referred to Manhattan Frame (MF). Given aset of inputs such as surface normals or vanishing points, we pose an MFestimation problem as a consensus set maximization that maximizes the number ofinliers over the rotation search space. Conventionally this problem can besolved by a branch-and-bound framework which mathematically guarantees globaloptimality. However, the computational time of the conventionalbranch-and-bound algorithms is rather far from real-time performance. In thispaper, we propose a novel bound computation method on an efficient measurementdomain for MF estimation, i.e., the extended Gaussian image (EGI). By relaxingthe original problem, we can compute the bounds in real-time performance, whilepreserving global optimality. Furthermore, we quantitatively and qualitativelydemonstrate the performance of the proposed method for various synthetic andreal-world data. We also show the versatility of our approach through threedifferent applications: extension to multiple MF estimation, videostabilization and line clustering.
展开▼
