Di Zhang, De Xu, Rui Song, Chaoqun Wang, and Yinchuan Wang
3D vision, feature extraction, visual SLAM, line feature, vanishing point, reprojection error
Visual simultaneous localisation and mapping (SLAM) estimates camera’s trajectories predominantly relied on point features. However, point features based methods have high requirements on the quality and quantity of point features, and sometimes the robustness of the algorithm is not enough in low-texture environments with changing illuminance. Moreover, the sparse map composed of point features reflects less environmental information. In this paper, a robust efficient monocular visual SLAM system is developed, making use of both point features and line features to compensate the weaknesses of traditional methods only based on point features. However, degeneracy occurs more severely when using line features rather than point features and causes the inaccuracy of the line features’ reconstruction. And the popular measurement model of line feature used in optimisation cannot correct the direction vector in the Pl¨ucker coordinates sufficiently because only the normal vector is used to calculate reprojection error. The reason of line’s degeneracy is also analysed, and a novel cost function using vanishing points (VPs) obtained from the images is also proposed to avoid this problem and correct the direction vectors of the line features. Finally, we conduct experiments using public datasets to compare the proposed method to the state-of-the-art algorithms, and the results indicate that the proposed method has better performance in mapping and estimated trajectory accuracy, which verify the effectiveness.
Important Links:
Go Back
