The goal of visual tracking and estimation of the location of feature points in the local environment is to use those point features as landmarks for pose estimation for a robot. The problem is dual to the mapping problem discussed earlier. In the mapping step, the azimuth to a point feature and our location are concatenated to produce the measurement, and the location of the point feature is the state vector which is to be estimated. In the localization problem, the azimuth to the point features and the landmark positions are concatenated to produce the measurement vector, and the vehicle pose is the state vector which is to be estimated. An EKF filter approach is again used to recursively improve estimates of the rover pose given new measurements.
In the above figure, we have three landmarks which have locations that have been estimated to within some uncertainty. For the result shown, the rover had an extremely large initial uncertainty, causing the EKF to virtually ignore the initial state vector. Using a system of nonlinear equations which geometrically constrain the rover location given the nominal landmark positions and azimuth measurements, the EKF determines a new estimate for the rover position. By combining dead reckoning errors, which monotonically increase with traverse distance, with the estimates given by the landmark observations, we can improve overall pose estimation.
Future work in estimating the position of observed landmarks includes:
 |
This page is maintained by Matthew Deans, a
robograd
in the Carnegie Mellon University School of Computer Science.
Comments? Questions? Mail me at deano@ri.cmu.edu Last Modified November 18th, 1997 |