An Improved ORB-SLAM Algorithm for Mobile Robots
| Autor: | Xiaolin Ma, Xiaodan Wang, Hailan Kuang, Xinhua Liu, Linjun Chen |
|---|---|
| Rok vydání: | 2019 |
| Předmět: |
0209 industrial biotechnology
Matching (graph theory) business.industry Computer science Feature extraction Mobile robot 02 engineering and technology Simultaneous localization and mapping 021001 nanoscience & nanotechnology Scale space Octree 020901 industrial engineering & automation Region of interest Obstacle avoidance Trajectory Computer vision Artificial intelligence 0210 nano-technology business Orb (optics) |
| Zdroj: | 2019 Chinese Control And Decision Conference (CCDC). |
| Popis: | In the ORB-SLAM system for mobile robots, there are many problems such as large matching error, slow operation speed, low positioning accuracy and small map application scope. To solve these problems, this paper first adds the depth information based on saliency detection and preprocessing the scene images to improve the efficiency of the system. Then, the ORB feature extraction is carried out in the scale space with large isolation. The improved multi-probe LSH and PROSAC algorithm were used to optimize the matching strategy and it will improve the matching accuracy and efficiency. Aiming at the large number of error closed-loop in the closed-loop detection algorithm, an improved closed-loop detection algorithm based on the region of interest of scene image and the idea of hierarchical weighted matching is proposed to improve the accuracy and recall of closed loop detection. Finally, the final motion trajectory and 3D environmental map are obtained through the pose image optimization and the octree building. The experimental results show that the method can effectively improve the positioning accuracy and computation efficiency. At the same time, the octree map can not only greatly save a lot of storage space, but also meet the following requirements such as navigation, obstacle avoidance and interaction. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|