使用部份區域特徵解決遮蔽物件的辨識系統

Abstract

在一般的生活情形下，物體遮蔽的問題總是發生在日常生活中，總是不易得到一完整無缺的物件影像，在只得到遮蔽物件的情況下，本論文利用2D影像資訊為基礎解決物遮蔽的辨識問題，進而了解受遮蔽的部份。此一系統利用混合高斯機率模型建構背景模型，對目前影像做背景濾除以取得前景，並利用簡單的邊緣偵測法找出物件的輪廓當其重要的資訊。在得到輪廓的資訊後，為了得到辨識遮蔽物件的能力，必須將輪廓的完整資訊分割成多塊輪廓的部份資訊，如此便可克服部份輪廓被遮蔽所受的影響。在此分析一些切割技術的理論與作法以及各方式的優缺點，以及說明所選用的非參數顯著點偵測法經過改進後的效果。在物件辨識系統的辨識階段所採用的方法，由於已經對輪廓進行切割，所以不再是一對一的做完整比對，而是要對每一物件所切出來的多個特徵做階層式的比對。在辨識出正確物件後再進而找出哪部份被遮蔽。最後以實驗結果說明此系統的效能，以及探討其優缺點和該改進的地方。

In this thesis, an occluded object recognition system based on partial shape feature is proposed and implemented. The foreground image of object is acquired by the background model built by Gaussian Mixture Model method. We applied the simple Edge Detection to obtain the contour of the foreground as the input of the occluded object recognition system. In order to recognize the occluded object, we must split the complete contour to many partial contours. Hence we will overcome the effect of occlusion. We analyze some split technology and compare the advantage and drawback. We propose an improved non-parameter dominant point detection system and experiment. In the recognition stage, we will use hierarchical compared system. After recognize the correct object, we try to find the occluded part of object. Lastly, we list the results of experiments and discuss the advance of the system.