Monocular motion estimation using a long sequence of noisy images

A kinematic model based approach is discussed for the estimation of 3-D motion and structure parameters from a sequence of noisy monocular images. The approach is based on representing the constant velocity translation and constant angular velocity motion using nine rectilinear motion parameters, which are the 3-D vectors of initial position, linear velocity, and angular velocity. The rotational motion is propagated in the kinematic model using the standard 3 × 3 rotation matrix. The measurements are noisy perturbations of 2-D image locations of feature points. It is assumed that the 3-D feature points are extracted from the images and matched over the frames. The structure of the moving object is represented by the coordinates of feature points in a 3-D coordinate system fixed on the object. A nonlinear least squares method is used to formulate the batch estimation of motion and structure parameters.