Hybrid control for mobile target localization with stereo vision
In this paper, we control image collection for a mobile stereo camera that is actively localizing a group of mobile targets. In particular, assuming that at least one pair of stereo images of the targets is available, we propose a novel approach to control the rotation and translation of the stereo camera so that the next observation of the targets will minimize their localization uncertainty. We call this problem the Next-Best-View problem for mobile targets (mNBV). The advantage of using a stereo camera is that, using triangulation, the two simultaneous images taken by the robot during a single observation can yield range and bearing measurements of the targets, as well as their uncertainty. A Kalman filter fuses the full state history and covariance estimates, as more measurements are acquired. Our solution to the mNBV problem determines the relative transformations between camera and targets that will minimize the fused uncertainty of the targets' locations. We determine a motion plan that realizes the mNBV while respecting field of view constraints. In particular, with every new observation, we compute a new mNBV in the frame relative to the camera and subsequently realize this view in global coordinates via a gradient descent algorithm that also respects field of view constraints. Integration of mNBV with motion planning results in a hybrid system, which we illustrate in computer simulations. ©2013 IEEE.
