| Summary: | 碩士 === 國立成功大學 === 電機工程學系碩博士班 === 92 === How to detect and track moving objects in video streams is an important and challenging research problem in visual tracking applications. Among the moving object detection algorithms, the background subtraction method has been widely used due to its less computational load and high detection quality. Nevertheless, with problems such as “sensitive to the changes of lighting” and “background initialization”, the background subtraction method often performs poorly in complex environments. To overcome this difficulty, a modified adaptive background subtraction method that can dynamically update the background model is proposed. On the other hand, the SSD method is a popular image tracking technique in visual servoing applications. However, the SSD method is very sensitive to the changes of target appearance. As a consequence, it may experience failures when applied to realistic environments. To overcome this difficulty, a multi-cue template matching algorithm that consists of several kinds of tracking algorithms is proposed to improve the practicability and robustness of the visual tracking system. To evaluate the performance of the proposed approach, the modified adaptive background subtraction method and the multi-cue template matching algorithm are applied to the real-time visual tracking system developed in our laboratory. Experimental results show that the proposed approach exhibits satisfactory performances.
|
|---|
