Visual Target Motion Analysis Using Controlled Camera Motion

We describe a set of techniques whereby absolute position and velocity of a moving target can be obtained with the aid of a position controlled visual sensor. We investigate how the sensor trajectory can be optimized to reduce the sensitivity of the target parameter estimates to noise in the measurements. We present a discrete time technique for obtaining the depth and motion of a moving target from target bearing measurements obtained at four viewpoints. We also present a recursive estimation scheme for obtaining the target position and motion parameters from a time-sequence of target bearing measurements. We present a technique for locally optimizing the camera position and orientation by moving the camera in the direction of the gradient of the trace of the estimation error covariance matrix. The algorithm produces trajectories which generally move toward the target with small oscillatory side to side motions superimposed. If we assume a foveal sensor model, the algorithm causes xation of the target.