Sensitivity study: The effects of beacon location errors on a vehicle's position and attitude estimation for a vision-based navigation system
Haas, Brian Michael
MetadataShow full item record
A sensitivity study is conducted for this thesis work where the effects of beacon location errors on the estimation of a vehicle's position and attitude are examined. Beacon location variation is introduced into the standard vision-based navigation (VISNAV) problem as zero-mean Gaussian noise. Two different simulations are carried out to study the effects of this noise on the estimation of a vehicle's position and attitude. The first simulation incorporates the Levenberg-Marquardt (LM) algorithm and the Monte Carlo method together to study the effects of beacon location variation on position and attitude estimation. The second simulation is time-based in nature and incorporates both the LM algorithm and the Gaussian Least Squares Differential-Correction (GLSDC) algorithm together to estimate both position and attitude for two different geometrical situations. A derivation of a measurement-error covariance matrix and incorporation of this covariance matrix into the estimation routine through the weighting matrix is presented in this thesis. The measurement-error covariance matrix accounts for the nonlinear noise addition of the beacon location variation for the measurement model of the VISNAV system. Simulation results show that incorporating the measurement-error covariance matrix into the estimation routine improves the vehicle's position and attitude estimates for the Monte Carlo simulation and the time simulations. The study looks at the calculation of the measurement-error covariance matrix using the state estimates instead of the true values for the simulation. Results show that there are no significant differences with calculating the measurement-error covariance matrix using the state estimates or the true values. This valuable result shows that the measurement-error covariance matrix can be employed in the estimation algorithm for a real-time system where the vehicle's true position and attitude are not known and the only valuable information available is the state estimates.