在大部分電腦視覺應用領域例如:影音監視、交通監控、人類移動擷取以及人機介面的互動裡，在某個場景中前景物件偵測通常被稱作為背景相減並且是一個關鍵性前處理的步驟。背景相減是一個廣泛被使用於從現時畫面與參考畫面的差異所得到的移動物件偵測的方法，其中參考畫面又稱之為背景影像或背景模型。在基本的原則裡，背景影像必定表示一張無移動物件之景像並且保持有規則的背景更新於照明的變化條件及在介紹裡所提及的一些問題。因此如何維持一張背景影像是非常重要的議題。 在此論文中，為了獲得含有以上所述的一些問題的精確之前景物件偵測，一個多模型背景維持演算法被提出。此多模型背景維持的架構包含兩個主要特徵去重建一張實際含有時間變化背景改變之背景影像。在這個架構下，此背景影像由每個像素最具意義、一再發生的特徵與主要特徵所表示。主要特徵由靜態與動態特徵組成來表示背景像素。此多模型背景維持包含兩個主要的步驟:背景維持與前景萃取。實驗顯示提出之方法在不同的連續畫面提供好的結果。量化評估及比較其已存在的方法顯示提出之方法提供改善較佳的結果且具有較低的運算複雜度。最後我們使用IEKC64x平台去實現多模型背景維持演算法來獲得即時前景物件偵測。 Foreground object detection in a scene, often referred to as “background subtraction”, is a critical early in step in most computer vision applications in domains such as video surveillance, traffic monitoring, human motion capture and human-computer interaction. Background subtraction is a widely used approach for detecting moving objects from the difference between the current frame and a reference frame, often called the “background image”, or “background model”. As a basic, the background image must be a representation of the scene with no moving objects and must be kept regularly updated so as to adapt to the varying luminance conditions and some problems described in the introduction. For this reason, how to maintain a background image is very important issue. In the thesis, in order to acquire accurate foreground object detection with above some problems, a Multi-model Background Maintenance (MBM) algorithm is proposed. A MBM framework contains two principal features to construct a practice background image with time-varying background changes. Under this framework, the background image is represented by the most significant and recurrent features, the principal features at each pixel. Principal features consist of static and dynamic features to represent background pixels. A MBM includes two major procedures, background maintenance and foreground extraction. Experiments show proposed method provides good results on different kinds of sequences. Quantitative evaluation and comparison with the existing method show that the proposed method provides much improved results with lower complexity. Finally, we use IEKC64x platform to implement MBM algorithm for obtaining real time foreground object detection.
