Contour-SLAM: A Robust Object-Level SLAM Based on Contour Alignment
Shiqi Lin, Jikai Wang, Meng Xu, Hao Zhao, Zonghai Chen
Abstract
Objects in environments are more stable than feature points, which can benefit for building a wide range of common view relationship and effectively reduce the cumulative error of visual localization. In this article, we propose contour-SLAM, the first RGB-D object-level SLAM method that uses the projection constraints between the dense object model and its instance masks for state estimation. First, considering that the unstructured object models are difficult to describe mathematically and their reprojection errors are difficult to quantify, we use voxels and cuboids to model objects and define the reprojection error between the 3-D object model and its 2-D instance segmentation mask based on the outer contour distance. Second, we construct a covisibility graph between keyframes based on the shared observations of same feature points and objects, which is further used to maintain a local map including keyframes, map points, and object models in real-time. Finally, a multiview bundle adjustment (BA) formulation is proposed to jointly optimize the three components of the local map. Experimental results on public dataset and our own collected sequences demonstrate that object modeling and camera pose estimation can benefit each other. The proposed method demonstrates competitive performance compared with the state-of-the-art SLAM methods.