Pas Ywe Xilt?gation

.2 rntthod IF prcpctc-d for dctc~rniuing the motion of a b ~ d y rclativc to a fixcd rn.tirr:nrnt.nt using thr rhangmg image seen by a camcra attached to thrd Imly. The cq~ttcz! flo-c. in t h e Image plane is t h e input, while the instants~;eous rotation and transIatlon of t h e body are t h e ~ i i t p i t t . if oj)tlcnl -flow could be ctclertn~tted p ~ e c i w l y , ~t w.voutd only h;v~ to be h o v n a t a few places to compute t l ~ r parameters of the n$otion. In practice, hwvever, the ~neasured optical flow will be somec% hat inaccural e. It is therei,jre advautageous t o c-omider methods sb ich use as much of the avallablt. information as postiblc. IVo cmpluy a Icast-squarcs approach which minirnizcs soxnc rncasurc of the disrrcpancy between t h c mcasured How and that predlctrd from thc comput~ )d motion parametex. Several diFerent error norms are investigated. In general, our algorithm lead.: to a s y s t ~ m of nonlinear equation? from xhictl t h e m o t ~ o n parameters may be ::)ml!uteci ~~rlrnerically. flowever, in t h e special c a m uheie f l i t 1 motion of the camera 1s purely t ra~i~iat lonwl or purely rotat lonnl , u i e c,f t h e ttppropr~,ztz !lorn, leads t o ;t FJ stem ol equallous f r o ~ r ~ ;zhich these parar~wters call be determmtd iu closed form.