Low-cost Unmanned Aerial Vehicle have large potential for applications in the civil sector. Cheap inertial sensor alone can not provide the degree of accuracy required for control and navigation of the UAVs. Additional sensors, and sensor fusion techniques are needed to reduce the state estimation error. In the context of this work, the possibility of using visual information to augment the inertial data is investigated. The projected image of the ground plane is processed by a homography constrained optical flow description and used as a measurement update for an EKF (Extended Kalman Filter). A simulation environment is used, which models the UAV dynamics, the rate-gyro and accelerometer errors as well as a low resolution vision system. The results of the simulation are very promising, as they show that the combination of the visual information and the inertial sensors can for particular situations perform better by a factor of 20 when compared to simple integration of low-cost inertial sensors. Previous works using similar methods did not use the projection matrix directly as a measurement. They extracted the state information from the matrix and relied on third sensors to solve the ambiguity in the velocity and height data. The proposed approach is able to handle this situation without the need for additional reference information and is stable even if the state estimate is inaccurate. The proposed method uses the constant brightness constraint method for relating consecutive images. Another possible solution is to perform feature extraction and relate the features of consecutive images. For this approach to work, the image requires enough visual information so that features can accurately be identified. A combination of the two approaches could potentially solve a wider class of problems. The sensor fusion method applied in the context of this work uses the EKF, which has known problems with highly non-linear systems. An alternative approach which might have advantages in this area is the class of Unscented Kalman Filters.
Homography based state estimation for aerial robots
Homographie-basierte Zustandsschätzung für Luftfahrt-Roboter
2008
8 Seiten, 2 Bilder, 15 Quellen
Conference paper
English
Homography-Based State Estimation for Autonomous UAV Landing (AIAA 2017-0673)
British Library Conference Proceedings | 2017
|Robust Homography Estimation from Planar Contours Based on Convexity
British Library Conference Proceedings | 2006
|Infinite Homography Estimation Using Two Arbitrary Planar Rectangles
British Library Conference Proceedings | 2006
|