Exploratory motion generation for monocular vision-based target localization

This paper presents a trajectory generation algorithm for the exploratory motion of a monocular vision-based target localization system. The planning algorithm uses the Fisher Information Matrix (FIM) to quantify the information presented to the localization subsystem. Maximizing functions of the FIM minimize the best achievable error bounds on the target estimate. Current approaches solve this optimization process by assuming the object location is known. This assumption is paradoxical as the goal of the overall system is to estimate the location of the unknown object. Therefore, an iterative approach has been developed. Furthermore, current approaches maximize the information content for a given path length or time. The new algorithm presented here minimizes the path length necessary to achieve a specified level of information while incorporating other motion constraints such as range limits, viewing angles, and safety margins based on the predicted estimate covariance.