Next:Recognizing Volumetric Models in Up:Active Perception Previous:Active Perception

Active Vision Using Optimal Control

K. Benameur, P.R. Bélanger

In this project, we used Optimal Control Theory to calculate trajectories that are best from an estimation point of view. We start from a dynamic model of an object whose position and velocity coordinates are to be estimated. Camera observations are seen to be 2D projections of a number of object feature points on a plane; the image points are dependent in a known manner on object and camera locations, i.e. are related to these by known equations. These observation equations, and the observation noise, depend on such variables as zoom and position. It is possible to formulate an optimization problem where the control variables are the forces acting on the cameras, and the performance index is a function of the covariance of the errors in the estimates generated by a Kalman filter. The outcome is a trajectory designed for optimum estimation of object position and velocity. We have used the results also to study the grasping of a moving object, where a compromise is needed between control and estimation.

Annual Report

Fri Nov 26 23:00:32 GMT 1999