Vision-aided IMU Odometer Using Geometric Constraint and Improved ORB Features
| Autor: | Hung, Chiao-Hsuan, 洪巧瑄 |
|---|---|
| Rok vydání: | 2018 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 106 Many methods based on sensor fusion have been proposed to minimize the error accumulation induced by the inertial measurement unit (IMU). Combining cameras and IMU together is an effective way to get a reliable estimation of ego-motion since the accumulation error from IMU can be constrained by the information of camera images. Meanwhile, IMU can provide real scale which a single camera lacks. In the proposed visual-inertial odometer algorithm, the multi-state constraint Kalman-filter and the trifocal tensor geometry are used to achieve sensor fusion. Also, a RANSAC algorithm with three view geometry constraint can avoid feature points which are mismatched or located on the moving object. For real-time implementation, feature detection and matching are the most time consuming procedure. Therefore, an improved ORB feature extracting method is proposed and used to reduce the dependency of hardware resources. Compared with the traditional ORB, the proposed method which utilizes rotation information of IMU is more suitable for tracking features in multiple views. On the other hand, the line information is used in visual geometry constraints as well. The experiment of the feature tracking includes repeated pattern problem, low texture problem and clutter texture problem. Besides, a hardware which integrates a 10-degree-of-freedom IMU, a monocular camera and a GPS receiver is used to validate the trajectory estimation method in real environment. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|