Scale Recovery Using Vanishing Points for Indoor Mobile Vehicles With Monocular Vision

This paper proposes a scale recovery method for indoor mobile vehicles with monocular vision. We aim to solve the scale ambiguity problem which is the main weakness of monocular SLAM. The method is based on the detection and modeling of the ground plane under the constant camera height assumption which is suitable for ground mobile vehicles. The ground plane detection relies on vanishing points estimation, and the ground plane modeling is accomplished by point clouds on the ground complemented with vanishing directions estimation. The method is evaluated both on the self-collected indoor dataset and the KITTI benchmark dataset. The experimental results verify the effectiveness of our scale recovery method.