Robust Orientation Estimation from Only A Plane and A Line in Manhattan World

The orientation estimation is an essential issue for indoor robotic tasks. Existing methods of orientation estimation in indoor environments suffer from poor accuracy, poor robustness and heavy computational burden due to drift error, textureless and clutter environment, and exhaustive search strategy. To overcome these deficiencies, a novel method of orientation estimation from only a plane and a line under the assumption of Manhattan World is proposed. Firstly, the framework of USAC is extended to estimate the dominant plane robustly. Then, the one plane and one line USAC is proposed to perform orientation estimation. Finally, we design a new cost function to refine the orientation estimation via nonlinear optimization. Extensive experiments show that the proposed method outperforms the state-of-the-art methods.