以適應性KLT為特徵之移動物定位系統

Abstract

近年來隨著科技一日千里、急速的發展成長，結合人們生活的科技產品如十字路口的監控系統，保全居家警示系統，GPS導航系統等不斷的推陳出新，而追蹤的效能就是這些系統能穩定執行的前提。追蹤問題已經是近十年來，許多科學家致力於研究的方向，但卻有許多無法克服的困難，如遮蔽、環境變化等，科學家提出許許多多的方法，但卻沒有一套追蹤系統，能完整的解決這些問題。如何將人類的視覺，透過攝影機轉化給電腦分析，讓電腦能有著人體的視覺感受，這就是追蹤的核心價值。 本研究所提出的追蹤系統結合了梯度與學習兩項重要的基礎，改進了當今許多研究只利用其中一類可能會產生的問題。以及本研究提出之方法，與許多現今方法不同點在於，本研究提出之系統，不需要建立背景，可隨插即用；並處理了許多現階段其他追蹤系統無法解決的問題。本篇論文提出一偵測與追蹤系統，其具有先進的適應性與即時的處理效率。

With accelerated advances in technology in the recent years, more and more new products are invented to increase convenience in our daily life, such as the crossroads monitoring system, home security alarm system and GPS navigation systems. Stable and robust tracking performance is the fundamental requirements for these systems to provide satisfactory and reliable functions. Over the past decade, tracking problems become the focus of many researchers; however, there still are many insurmountable difficulties remained such as shelter, environmental changes. Various temporary solutions are provided to tackle these issues, but without a complete tracking system, the achievement might only be imprudent and partial. How to turn human vision into computer vision via simple cameras, so the computer can have a human visual experience, is the core value of the tracking problem. Unlike the previous researches, the tracking mechanism proposed in this study combined gradients and learning. And the plug and play system can operate immediately after installation without any background prerequisites, are proved to be able to deal with many other problems that others are unable. In this thesis, a detection and tracking systems are introduced to provide advanced, adaptive and real-time processing efficiency.