AN ITERATED ESTIMATION OF THE MOTION PARAMETERS OF A RIGID BODY FROM NOISY DISPLACEMENT VECTORS

An important and still controversial problem is the robust recovery of the parameters of motion of a rigid body from noisy point matches made on perspective views of the body. In this correspondence, we formulate a minimization problem which permits relatively robust parameter estimation and helps overcome poor zoom estimation when the field-of-view is small. We then assume that the motion has a small rotary component, interframe differences are small, and the errors in the system are due to independently distributed errors in the components of the displacement vectors. A fast procedure for minimization is exposited, in which the parameter set is partitioned and conditional generalized least-squares formulas identified. Recursive application of these provide the search space for the minimization problem. Comparative results are presented using simulated data and using displacement vectors obtained from an intensity-based matching algorithm. © 1990 IEEE