The use of image plane velocity measurements in recursive 3-D motion estimation from a monocular image sequence

A recursive algorithm for estimating the three-dimensional trajectory and structure of a moving rigid object in an image sequence has been previously developed. The algorithm uses the location of a set of tracked feature point positions as input to a nonlinear extended Kalman filter. This paper examines the incorporation of image plane velocity measurements into the same recursive framework. It is shown that the incorporation of image velocity information significantly improves trajectory estimation as long as the relative error in velocity measurements is less than about 160%. It is also shown that using image velocity significantly improves the estimator's tracking performance during filter initialisation as well as during abrupt changes in the 3-D motion model.<
>