Passive Navigation in a partially known environment

The paper presents an integrated solution to the problem of obtaining the kinematics of a moving vehicle and the 3D locations of salient points in the external environment, based on a sequence of monocular images. The structure of a small number of navigational landmarks in the scene is assumed to be known. The parameters of interest are expressed in an inertial, or 'world' coordinate system, external to the moving vehicle, and are estimated recursively using an extended Kalman filter (EKF). The extraction and matching of feature points are done using a gabor wavelet representation, and are interleaved with the recursive estimation. Experimental results on a real image sequence are given.<>