Image Tracking Analysis and Simulation

The problem considered herein is that of tracking a well-resolved object with an active direct detection sensor system. The sensed object image is corrupted by additive as well as multiplicative (speckle) noise. A steady-state object tracking filter based on a discounted weighted least-squares procedure is derived in closed-form, along with its error covariance properties. The filter, which is of a polynomial-in-time type, estimates the position, velocity, and acceleration of an object moving in space, based on noisy measurements of position offset in the focal plane array. The position offset measurements are obtained using the Fitts correlation algorithm. The error covariance matrix of the Fitts algorithm is derived in closed-form for both additive and multiplicative noise. The discounted weighted least-squares filter has the advantage of requiring only a single parameter, the discount factor, as opposed to three parameters in the conventional a-0-y filter. Most importantly, the discount factor guarantees a certain degree of stability in the filter. Simulation results are given for the case of a computer-generated speckled image. Tracking performance results are presented for the discounted weighted least-squares filter, the a-0-y filter, and the recursive Kalman filter, using the Fitts correlation algorithm for generating position offset measurements in all cases.