Abstract
A robust computer vision system for an unmanned surface vehicle (USV) is being developed, in support of tracking a moving marine vehicle from the USV. A methodology to minimize the tracking errors regarding the raw 2-D measurements of the position of a target boat using a commercial stereo camera will be presented. This errors are mainly due to image quantization limitations and pixel miscorrespondences in the stereo-matching process. While more sophisticated matching algorithms may lead to a better depth reconstruction of the scene at a high computational cost, simple matching algorithms perform faster at the expense of larger error in depth measurement. This approach consists of combining a simple stereo matching algorithm, along with an extended Kalman filter (EKF) in the time domain, as compared with the image domain, leading to a lower algorithm complexity and less computational time, minimizing the errors of position measurements of a target boat in the context of marine navigation. Suitable choices of the measurement and motion models of the target boat are made in order to accomplish a satisfactory response of the system.