Observer trajectory generation for target-motion estimation using monocular vision

One important objective of many autonomous robotic missions is target-motion estimation. This task requires an autonomous observer vehicle to determine the global position and velocity of an object in the world, given only measurements of its own position and velocity and components of the relative position (either bearing or range) between it and the target object. One specific configuration of sensors uses a single camera in conjunction with navigation sensors such as GPS. The resultant sensor fusion problem which yields an estimate of the global target state is known as target-motion estimation using monocular vision.; One major complication with target-motion estimation using monocular vision is the dependence of the estimation performance on the specific observer path. Because the monocular vision system provides only a bearing measurement to the target, the camera must move to multiple locations in order to obtain a solution. The accuracy of the resulting target-state estimate is a function of the specific camera path, and therefore some trajectories lead to better performance than others. The goal of the work presented here is to enhance target-motion estimation using monocular vision by generating near-optimal observer trajectories.; This dissertation presents the details of a novel observer-trajectory generator that focuses on three important issues. First, the limited field of view of the vision system is addressed. Second, a new optimization objective is desired for use on operational vehicles. Third, the uncertain nature of the target-state estimate leads to a difficult conundrum—an optimal trajectory can be determined if the target motion is known; however the whole point of the trajectory generation is to enable the estimation of the target's unknown motion.; Important features of the new trajectory generator include the identification of a quality metric used to evaluate candidate trajectories, a novel trajectory-generation algorithm for known initial target state capable of quickly finding near-optimal paths, and integration of this algorithm as the core of a new trajectory generator for the general case. The success of the trajectory generator is verified by experimental results obtained using a micro autonomous rover platform.