Analysis of sensor movement errors in monocular vision-based tracking systems

Abstract

When a vision sensor is used to track an object in an outdoor realistic navigational environment, it is subjected to unexpected movements or vibrations of the mounting platform. In this paper, the performance of such a setup in terms of range and heading angle errors is studied. The noise generated by the navigational environment is represented in two ways: sensor rotation angle errors and image coordinates errors. A consistent detectable region is obtained such that the tracked object is always seen by the sensor. Based on this region, a reliable region consisting of no singularity points is defined so that the range error does not become infinity. The optimum values of a controllable subspace, consisting of the object height and depression angle, with respect to an uncontrollable subspace, consisting of object coordinates and sensor movement errors, are then found by employing the mini-max estimator for the worst case performance and the minimum mean-squared estimator for the average performance.<>