| Summary: | 博士 === 國立臺灣大學 === 電機工程學研究所 === 97 === The Bayesian filter based visual tracking and visual servoing systems are presented in this dissertation. In order to achieve the robust visual tracking performance in real-time, we extend some probabilistic methodologies, such as the probabilistic data association filter and the particle filter, to solve the problems generated by the vision sensor. We focus on target estimation with noisy measurements, which may be due to the mechanical noise arising from undesirable shaking of the camera, optical interference from similar objects, cluttered background, occlusions, and lighting changes, etc. We also try to overcome the target overlapping problem, which arises from loss of depth information of different targets with only 2-D visual observation.
The proposed visual tracking algorithms are then applied to real-time control of the active camera platforms to fulfill the visual servoing function. Along this line of research, visual servoing systems are particularly designed such that an unmanned aerial vehicle (UAV) chases another aerial vehicle, an UAV performs the mission of aerial reconnaissance, and a single moving camera simultaneously tracks multiple targets. Moreover, we utilize our designed visual tracking and visual servoing algorithms to develop several distributed vision systems. A wide area surveillance system is constructed to track multiple moving objects through effective cooperation of multiple distributed vision sub-systems, each with a pan-tilt camera.
|
|---|
