Study on object extraction from image sequence for target tracking

Abstract

Automatic video tracking systems are employed in a wide variety of applications such as surveillance system, fire control systems, guidance, robotics, and autonomous vehicle navigation. Especially, for the military applications, the development of automatic target tracking systems has enabled more accurate determination of target position, velocity, acceleration, and other parameters required for weapons guidance and target designation. In applications which require target designation and firing such as a battle tank or missile, it is important to determine the center of the target since the centroid of the target image is used for the aiming point of the fire control systems. For the hard targets like tanks, the determination of an exact aim point on the target is essential to achieve high probability of hit. For the computation of the exact aiming point, an accurate extraction of the target from the image is indispensable requirement. However, several problems may arise in the process of the object extraction. One of the major problems is noise and generally, two sources of noises degrade the performance of the object extraction. The first one is system and sensor noise, and it is usually modeled by additive white Gaussian noise and can be easily rejected. However, another source of noise, clutter which has similar intensities to those of the target, make it more difficult for the tracker to extract the target from its surrounding background. This problem brings about misclassifications between the target pixels and background ones, and causes critical errors in computing the aiming point. Thus, the tracking system often fails to keep the aiming point and produces unfavorable results in the outdoor environments. In this dissertation, new methods on the target extraction are presented in order to guarantee the performance for finding the reliable aim point against cluttered environments. The goal is to develop an algorithm which is robust in outdoor environme...