Scale-aware monocular visual-inertial pose estimation for aerial robots
| Autor: | Huimin Lu, Minjun Xiong, Junhao Xiao, Dan Xiong, Ming Lv |
|---|---|
| Rok vydání: | 2017 |
| Předmět: |
050210 logistics & transportation
0209 industrial biotechnology Monocular Computer science business.industry 05 social sciences Feature extraction ComputingMethodologies_IMAGEPROCESSINGANDCOMPUTERVISION Gyroscope 02 engineering and technology Simultaneous localization and mapping law.invention Extended Kalman filter 020901 industrial engineering & automation law Inertial measurement unit 0502 economics and business Computer vision Artificial intelligence Visual odometry business Monocular vision Pose |
| Zdroj: | 2017 Chinese Automation Congress (CAC). |
| DOI: | 10.1109/cac.2017.8244045 |
| Popis: | Monocular visual odometry algorithm has been widely used to estimate the pose of aerial robots in GPS denied environments. However, the pure visual system usually has poor robustness in large scale environments. This paper presents a pose estimation algorithm which fuses monocular visual and inertial data using the monocular ORB-SLAM algorithm as the visual framework. Firstly, the scale estimation algorithm is proposed to estimate the unknown scale parameter in a monocular visual framework based on IMU measurements, meanwhile the gyroscope bias of IMU and gravity direction are initialized. Then, the Extended Kalman Filter (EKF) algorithm is used to fuse the pose estimation results obtained by monocular vision system and IMU respectively. This method makes full use of the advantages of high-frequency characteristics of IMU and low drift of monocular vision system. Finally, the experiments are carried out on a small multi-rotor Unmanned Aerial Vehicle (UAV). The results show that the maximum error of position estimation is less than 1.326 meters, and the maximum error of attitude estimation is about 3.859 degrees, which verifies the accuracy, effectiveness and stability of the proposed algorithm. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|