טכניון מכון טכנולוגי לישראל
הטכניון מכון טכנולוגי לישראל - בית הספר ללימודי מוסמכים  
M.Sc Thesis
M.Sc StudentTomer Shtark
SubjectTracking a non-cooperative target using a real-time
stereovision-based control law
DepartmentDepartment of Aerospace Engineering
Supervisor Full Professor Gurfil Pinchas


Abstract

Tracking a non-cooperative target is an important challenge in the field of robotics, because in unfamiliar environments most targets are unknown and unspecified. Stereovision is suited to deal with this issue, because it allows robots to passively scan large areas and estimate the relative position, velocity and shape of the objects they recognize. In case the target moves, a control algorithm is needed in order to keep it within sight.


This research is an experimental effort aimed at developing, implementing and evaluating a real-time non-cooperative target tracking methods using stereovision measurements. The experiments were conducted in the Distributed Space Systems Laboratory at the Technion. A camera rig was built and installed on top of a robot, referred to as ``chaser''; visual data processing software and a control algorithm were developed.


A computer-vision feature detection and matching algorithm was used in order to identify and locate the target in the captured images. The visual data was sent to either an extended Kalman filter, an unscented Kalman filter or a converted measurement Kalman filter, which estimated the relative position and velocity. The quality and nature of these estimations were examined and compared to external reference data. A control algorithm was used for the purpose of keeping the target robot within the chaser's field of view.


Extensive analytical and numerical analyses were conducted on the multi-view stereo projection equations and their solutions, which were attained by using least squares and total least squares techniques.


The benefits of using the unscented Kalman filter and the total least squares technique are illustrated using an experimental and numerical evaluation. These findings offer a general and more accurate method of solving the static and dynamic stereovision triangulation problems.