| 英文摘要 | SLAM(Simultaneous Localization and Mapping), referring to the process that a mobile object localizes itself and simultaneously builds the 3D models of the unknown scene, is an important research topic in the field of computer vision. Recently, SLAM has been extensively applied into many applications including robot navigation, augmented reality, 3D reconstruction, etc. However, there still exist a few key problems for the SLAM technique, such as multi-sensor systemic calibration, dense mapping, loop closure detection, etc. In this thesis, addressing both the multi-sensor calibration and dense mapping problems, I explore a Visual-Inertial system, where a coarse-to-fine calibration method is proposed. Moreover, a visual SLAM system which can do dense mapping is developed. The main contributions are: (1) A coarse-to-fine calibration method is proposed, which is used to calibrate a Visual-Inertial system consisting of two cameras, one IMU and one inclinometer. The proposed method first calibrates the two cameras. Then with the obtained camera parameters, the relative pose between the IMU and a camera, as well as the relative pose between the inclinometer and a camera is computed initially. And the pose between the IMU and the inclinometer is calculated directly. The initially estimated relative pose between the IMU and the inclinometer is prone to be poor due to the limitation of the used motion platform, hence a new optimization algorithm, which is able to integrate high-quality measurements captured from the IMU and the inclinometer, is proposed to refine this relative pose. Extensive experimental results show the effectiveness of the proposed method. (2) A software package for calibrating the above Visual-Inertial system is developed, which integrates the coarse-to-fine calibration method with several modules and a friendly graphical user interface. The software package uses bottom-to-up design principle and divides the functionalities into several individual modules. The core modules can collect measurements from the two cameras, the IMU and the inclinometer simultaneously, control the 6 DoF motion platform, perform the camera calibration, calibrate the vision/inertial sensors, and optimize the calibration results. All the operations can be done through the graphical user interface, which makes the software package easy to use. (3) A monocular SLAM system is developed aiming the problem on dense mapping, which estimates dense depth by utilizing M... |
修改评论