3D Self-Localisation from Angle of Arrival Measurements.
Scientific Publication
- Report Number:
- DSTO-TR-2278
- Authors:
- Kim, J.; Hmam, H.
- Issue Date:
- 2009-04
- AR Number:
- AR-014-502
- Classification:
- Unclassified
- Report Type:
- Technical Report
- Division:
- Weapon Systems Division (WSD)
- Release Authority:
- Chief, Weapons Systems Division
- Task Sponsor:
- CWSD
- Task Number:
- 07/249
- File Number:
- 2008/1127664
- Pages:
- 29
- References:
- 7
- Terms:
- Optimisation
- URI:
- http://hdl.handle.net/1947/9898
Abstract
We propose a 3D self-localisation method that uses 3D angle of arrival (AOA) information (ie., azimuth and elevation measurements) from landmarks. The formulation is based on minimising the collinearity error between the estimated line of sight (LOS) to the landmark and the measured AOA. This method runs in two parts – initial estimation of the vehicle azimuth and position assuming the vehicle has no tilt, and iterative 3D pose estimation based on a small angle approximation approach. Simulation study indicates that this method is efficient, requiring a small number of iterations, globally convergent and robust.
Executive Summary
In this report, we introduce a 3D self-localisation method that uses three-dimensional angle of arrival (AOA) information (i.e. azimuth and elevation measurements) from landmarks. We assume here that such measurements are available, even though in practice they need to be processed from the onboard sensors such as cameras or RF receivers. This work is a follow on to our previous study on 2D localisation where the altitude and the tilt angle of the vehicle were not relevant. However, in many applications the vehicle on-board sensor may be tilted making bearing measurements erroneous. In these situations the localisation problem has to be formulated in six-degrees of freedom (i.e. 3 DOF for sensor position and 3 DOF for sensor orientation). The formulation is based on minimising the collinearity error between the estimated line of sight (LOS) and the measured AOA. Using such error metric, we arrive at an iterative algorithm that runs in two parts: initial estimation of position and azimuth assuming zero tilt, followed by iterative orientation and position updates using small angle approximation approach. The proposed method is evaluated against a benchmark known as the orthogonal iteration which is known to be accurate, globally convergent, and efficient. The experimental results indicate that the proposed method is more accurate than the benchmark. The proposed method is also efficient requiring a small number of iterations and appears to be globally convergent.