DPC-SLAM: Discrete Plane Constrained VSLAM for Intelligent Vehicle in Road Environment

Simultaneous Localization and Mapping (SLAM) plays a significant role in providing accurate localization for intelligent vehicles. However, the traditional six-degree-of-freedom motion pose estimation assumption does not align with the actual vehicle motion, leading to pose estimation errors. To address this problem, we propose a stereo-visual SLAM algorithm specifically designed for road scenes to incorporate road constraints into the system. Our method leverages road constraints to eliminate errors caused by the six-degree-of-freedom motion assumption. Firstly, we represent the road as a set of discrete planes, simplifying the complex road surface into discrete local planes, and establishing the constraint relationship between camera poses and local planes. Then, we propose a tightly-coupled SLAM system that optimizes camera poses, map points, and discrete planes simultaneously. Finally, we evaluate the proposed system on the KITTI dataset and our self-collected dataset. Experimental results demonstrate that our method significantly improves the accuracy and stability of camera pose estimation.