Vision Based Target-Tracking Realized with Mobile Robots using Extended Kalman Filter

The paper presents a novel scheme for target-tracking realized with two mobile robots, where one robot is configured as tracker and the other as moving target. Fuzzy C-means clustering algorithm has been employed here to segment the target robot in images grabbed by the tracker. A new localization algorithm has also been proposed to determine the location of the target in the segmented images. An extended Kalman filter has been employed here for predicting the direction of motion of the moving target from its current and last few positions. The robot is pre-trained with back-propagation learning algorithm to plan its trajectory amidst obstacles. The pre-trained neural net is used in target-tracking application to control the motion of the tracker in the predicted direction of the moving target. Performance of the proposed neuro-Kalman synergism in target-tracking has experimentally been found to be superior to a tracking scheme without prediction by Kalman filter.