Visual SLAM With Building Structure Lines

Simultaneous localization and mapping(SLAM) is one of the important researches for the mobile robot and monocular SLAM has become a hot topic in recent years. In this thesis a novel 6 Do F visual SLAM method based on the structure lines is proposed. The idea is that we use the building structure lines as features for localization and mapping. The man-made buildings exhibit strong structural regularity and in most cases can be abstracted as blocks that are stacked together with three dominant directions, which is known as Manhattan-world assumption. The lines which are aligned with the dominant directions are called structure lines. Unlike other line features, the building structure lines encode the global orientation information that constrains the heading of the camera over time, eliminating the orientation drift caused by the accumulation of angular errors and consequently reducing the position drift.In this thesis the background and related work of SLAM is discussed at first. Then some related computer vision theory and the EKF-SLAM theory are introduced. After this the algorithm of the proposed visual SLAM with building structure lines is presented in detail. We extend the standard EKF visual SLAM method to adopt the building structure lines with a novel parameterization method, a measurement model and a robust data association strategy. Finally a series of experiments have been performed for both synthetic and real-world scenes. 25 Monte Carlo experiments have been conducted for the synthetic scenes. The real-world experiments include a large scale scene from RAWSEEDS benchmark and a hand-held camera data. The results show that, compared with the state-of-the-art visual SLAM methods, our method yields no accumulated orientation error and remarkably less position errors. In the test of indoor scenes of the public RAWSEEDS datasets, our method produces bounded position errors about 0.79 m along a 967 m path, without applying any loop closing algorithms.