engleski

Stereo vision object tracking using Kalman filter estimation

In the course of this thesis, a vision based tracking solution was developed. The goal of the system was to track moving objects in a robotic environment. Basic projective geometry behind the stereo vision set is presented at the beginning of the thesis. The thesis also addresses the problems caused by vision system noise and provides solutions on how to solve them. The vision system extracts the object information based on color segmentation results. During the development of the tracking algorithm, two approaches were considered as solutions to the tracking problem. The original approach is based on a nearest neighbor algorithm. As this approach did not suffice in performance, the tracking algorithm was modi fied using an extended Kalman filter (EKF). Before the EKF was implemented a series of simulations were performed in Matlab/Simulink. This was needed for the EKF design process. Despite the problem with synchronizing the control system and the vision system, a valid solution was developed. The use of the EKF resulted in better performance than the nearest neighbor approach. The fi nal results showed that the use of EKF tracking algorithm is valid.

robot vision; object tracking; color segmentation; EKF; triangulation; epipolar geometry; nearest neighbour

nije evidentirano