自動化影像偵測系統在人形目標物之分析

Abstract

本文所提出的監視識別系統，利用新的影像分割技術取出入侵物並針對入侵物做人形分析。本系統與其他監視系統主要差異在本系統特別針對行人做分析。在以往傳統應用中架構下，影像只能做場景被動取得無法分析在螢幕中的重要目標。隨著時代的進步，電腦運算能力的發達使我們能即時地分析並做反映。這個監視識別系統由CCD camera對場景做影像擷取的工作和一部電腦做影像分析所組成的。系統包含三個主要的部分，第一個部分是尤拉數的選擇，第二個部分是影像分割技術，第三個部分則是人形分析。 動態偵測模組從連續影像中的差異來偵測移動物的出現。藉由stable Euler number technique讓電腦自動選定篩值。”Segmentation by Local Equalization”綜合Histogram Equalization和morphology operations的優點使得物件影像分割較單一篩選值分割法更趨完整。最後利用人體比例來分析入侵物是否可能為行人。本論文將以各個步驟實驗以證明系統正確性。

This thesis proposes an automatic image detection system that tracks intrude object and recognizes if it is a shape of pedestrian. The system has three main modules. The first is the stable Euler number technique. The second is the method of Segmentation by Local Equalization. The third is the module of pedestrian analysis. Motion detection extracts moving objects by difference between consecutive image frames. The system will decide the threshold by stable Euler number technique. The method of Segmentation by Local Equalization combines the advantage of histogram equalization and morphology operations to makes the shape of incursive object extracted more correct and complete. Finally, the system will analyze this object if it is the shape of walking human being.