Egocentric Indoor Localization from Room Layouts and Image Outer Corners

Egocentric indoor localization is an important issue for many in-home smart technologies. Room layouts have been used to characterize indoor scene images by a few typical space configurations defined by boundary lines and junctions, which are mostly detectable or inferable by deep learning methods. In this paper, we study camera pose estimation for egocentric indoor localization from room layouts that is cast as a PnL (Perspective-n-Line) problem. Specifically, image outer corners (IOCs), which are the intersecting points between image borders and room layout boundaries, are introduced to improve PnL optimization by involving additional auxiliary lines in an image. This leads to a new PnL-IOC algorithm where 3D correspondence estimation of IOCs are jointly solved with camera pose optimization in the iterative Gauss-Newton algorithm. Experiment results on both simulated and real images show the advantages of PnL-IOC on the accuracy and robustness of camera pose estimation over the existing PnL methods.