New sensing strategies for monitoring moving polyhedral objects by machine vision

A set of sensing strategies is proposed for monitoring three-dimensional moving objects by computer vision. Three-dimensional (3-D) object surface points are selected as the features for monitoring 3-D moving objects because the point features are easy to detects, extract, store, and manipulate. It is proved that the minimum measurable feature point set for monitoring a 3-D moving convex polyhedral object is exactly the set containing all the junction points of the objects. Based on the sampling theorem and several properties of photogrammetry it is proved that the minimum data-acquisition rate of a vision system monitoring 3-D moving objects can be determined with discretely sampled two-dimensional image sequence data alone. Certain properties of orthographic projection useful for determining the minimum number of sensors needed N_s to monitor 3-D moving convex polyhedral objects are investigated, and the bound on N_s are also derived. An algorithm for determining N_s and the corresponding directions of the sensors is proposed. The feasibility of the proposed algorithm is shown by three illustrative examples and an application example