Bayesian Estimation and Tracking: A Practical Guide

19.3 Components of the Observation Model

Assume that there are J cameras used to gather data on the state of the rigid body and that they observe feature points designated by markers placed around the surface of the rigid body. The location of each feature point relative to the body center of mass is known to within very tight tolerances due to prior calibrations. Let be a set of fixed 3D time-invariant locations of the feature points in the body coordinate system. Because each camera is placed so that it has a different field of view, each camera will view a different subset of markers. For the j th camera at time t_n, let the number of viewed feature points be K_j,n and let the individual visible feature points for that camera be designated so that each k_i,j,n refers to the point , where 1 ≤ k_i,j,n ≤ N_s. Assume that all of the image frames across all cameras are synchronized so that the cameras each record an image frame simultaneously every 0.005 s. Also assume that the position and orientation of all cameras is calibrated for every frame and that the association of marker images for each camera image frame with predictions of those marker image positions is managed manually. Figure ...