基於改良式獨立成分分析之人物偵測與追蹤

Abstract

近幾年來，人物偵測及追蹤在電腦視覺中是一項常被深入探討的領域，且其可被廣泛應用在居家照護、保全及病人監控等系統。本論文提出一改良式獨立成分分析技術的人形自動偵測系統。我們用獨立成入分析法抽取辨識特徵，且以條件熵來做為特微選擇的依據，以此得到具有良好辨識能力且具有代表性的特徵。強建的支持向量機則為我們系統中主要的數據分類法。我們的實驗環境包含室內及室外，而監視畫面中的移動物體則有行人、動物及車子等等。 我們使用背景相減法取出畫面中的移動物體。為了處理複雜背景的情況，使用高斯混合模型來建構背景。針對移動物體被部分遮避的情形，我們提出金字塔型橢圓形頭部偵測法來分離它們。此外，利用簡單的色彩資訊及卡爾曼濾波器進行移動物體的追蹤及動向預測。

In recent years, video based human detection and tracking are a popular research area, and it has been used in widely applications such as homecare, security, patient monitoring and so on. This paper introduces a human detection system using modified Independent Components Analysis (ICA). The ICA features are selected by conditional entropy and classified by Support Vector Machine (SVM). The proposed system monitors the movement of human, animals or vehicles which across a secured area, and it works well in indoor or outdoor environment. The background subtraction is used to extract moving objects. In order to handle situations where the background of the scene is cluttered and not completely static but contains small motion, we models the background based on Gaussian mixture model (GMM). In complex situation, the moving object may disappear totally and partially due to occlusion by other objects. A fitting ellipse function based modification pyramid method is used for separating some multi-person occlusion. Our system combines with Kalman filter to estimate motion information and use the information in predicting the appearance of targets in succeeding frames.