利用建立可信賴的背景完成影片中物體的分割與追蹤

Abstract

本論文提出一個能分割與追蹤影片中物體的演算法，我們提出了一種能建造出可信賴的背景的方法，利用未加工的背景可分離出移動的物體，再利用梯度向量流量(Gradient Vector Flow, GVF)的特性來分割移動物體的邊界，利用圖形主動輪廓 (Morphing Active Contours) 來做移動物體的追蹤，最後利用每個分割的結果建立影片的可信賴背景，本論文使得梯度向量流量(GVF)能在複雜背景下成功的分割目標物體，因為被分割出來的物體之面積僅是全影像的一部份，而我們只需要針對此部份做運算，因此可以減少部分的運算量，並因為建立了可信賴背景提昇了壓縮效率。 在本論文中做了多個真實影像的實驗，經由實驗結果知道，若物體呈像面積夠大而且移動速度不算太快下，所提的方法可以有效的分割與追蹤影片中之目標物體。

An algorithm of video segmentation and tracking based on background construction is proposed. Rough background can be constructed by using this algorithm. Applying rough background, the moving object can be segmented by using the characteristic of gradient vector flow (GVF). Morphing active contours is used to track the object that has been segmented. Finally, reliable background is constructed from the every segmentation result. In this thesis, the abilities of GVF in the complicate background have been improved by using construct reliable background. According the rough background, only the smaller area of moving object is used to segmentation. The area of tracking is reduced by motion estimation and the coding efficiency is improved because the construction of reliable background. In this thesis, several real videos are experimented: Under the assumption of the size and speed of object, efficient segmentation and tracking on multi-objects can be implemented by the algorithm proposed in this thesis.