| Formatted contents note |
An adaptive edge-enhanced correlation based robust and real-time visual tracking<br/>framework, and two machine vision systems based on the framework are proposed.<br/>The visual tracking algorithm can track any object of interest in a video acquired from<br/>a stationary or moving camera. It can handle the real-world problems, such as noise,<br/>clutter, occlusion, uneven illumination, varying appearance' orientation, scale, and<br/>velocity of the maneuvering object, and object fading and obscuration in low contrast<br/>video at various zoom levels. The proposed machine vision systems are an active<br/>camera tracking system and a vision based system for a UGV (unmanned ground<br/>vehicle) to handle a road intersection.<br/>The core of the proposed visual hacking framework is an Edge Enhanced<br/>Back-propagation neural-network Controlled Fast Normalized Correlation (EEBCFNC),<br/>which makes the object localization stage efficient and robust to noiseo<br/>object fading, obscuration, and uneven illumination The incorrect template<br/>initialization and template-drift problems of the traditional correlation tracker are<br/>handled by a best-match rectangle adiustment algorithm. The varying appearance of<br/>the object and the short-term neighboring clutter are addressed by a"robust templateupdating<br/>scheme. The background clutter and varying velocity of the object are<br/>handled by looking for the object only in a $tnamically resizable search window, in<br/>which the likelihood of the presence of the object is high. The search window is<br/>created using the prediction and the prediction error of a Kalman filter. The effect of<br/>the longierm neighboring clutter is reduced by weighting the template pixels using a<br/>2D Gaussian weighting window with adaptlve standard deviation parameters. The<br/>occlusion is addressed by a data association technique. The varying scale of the obiect , correlating the search u'indou uith three scales of the template, and<br/>'-::ptinu the best-match region that produces the highest peak in the three correlation<br/>i*r-.es. The proposed visual tracking algorithm is compared with the traditional<br/>- :relation tracker and, in some cases, with the mean-shift and the condensation<br/>.::;kers on real-rvorld imagery. The proposed algorithm outperforms them in<br/>:,-'tustness and executes at the speed of 25 to 75 frames/second depending on the<br/>JLirrent sizes of the adaptive template and the dynamic search window.<br/>The proposed active camera tracking system can be used to get the target<br/>ahrars in focus (i.e. in the center of the video frame) regardless of the motion of the<br/>tarset in the scene. It feeds the target coordinates estimated by the visual tracking<br/>tiameu'ork into a predictive open-loop car-following control (POL-CFC) algorithm<br/>rihich in turn generates the precise control signals for the pan-tilt motion of the<br/>camera. The performance analysis of the system shows that its percent overshoot, rise<br/>time. and mcmimum steady state error are 0%o,1.7 second, and +1 pixel, respectively.<br/>The hardware of the proposed vision based system, that enables a UGV to<br/>handle a road intersection, consists of three on-board computers and three cameras<br/>(mounted on top of the UGV) looking towards the other three roads merging at the<br/>intersection. The software in each computer consists of a vehicle detector, the<br/>proposed tracker, and a finite state machine model (FSM) of the traffic. The<br/>information from the three FSMs is combined to make an autonomous decision<br/>whether it is safe for the UGV to cross the intersection or not. The results of the actual<br/>UGV experiments are provided to validate the robustness of the proposed system.<br/>Index terms - visual tracking, adaptive edge-enhanced correlation, active camera,<br/>unmanned ground vehicle. |
