The usefulness of imaging sensors in motion prediction

It is shown that the response of an image-enhanced tracking loop is superior to that attainable from conventional trackers without imaging capability. Because of the non-Gaussian nature of the problem and the nonlinearities in the motion model, it is difficult to make inclusive generalizations about the relative performance. It is evident though that an imager may help to achieve good quiescent performance without a severe tuning penalty. This exchange is difficult to make with conventional algorithms. Two modified linear Gauss Markov procedures are shown to have similar performance, and in this application, increasing the order of the estimator to shape the ersatz Gaussian disturbance is not worth the additional complexity. The image-based architecture provides 40% better quiescent performance with a somewhat less pronounced improvement during a turn