自動態影像中偵測與追蹤移動目標物是智慧型監控系統必須具備的基本功能之一,但若能發展出可以同時偵測與追蹤多個目標物的視覺系統,將使監控系統的功能更完善。因此,本文的研究目的即在建構可以同時追蹤二個移動目標物的立體視覺伺服系統,此系統由二組立體視覺平台所構成,每一個平台包含二部攝影機及二個旋轉自由度。首先利用影像處理及多重閥值顏色篩選方法擷取動態影像中的移動目標物;接著,由於支持向量機器可以分類非線性不可分離的資料類型,因此本文建立一支持向量機器用以辨識移動目標物;最後找出目標物的質心,再運用立體視覺演算法計算目標物與攝影機之間的相對位置,將此相對位置轉換成馬達控制訊號以使立體視覺平台能追蹤各自的移動目標物並將之鎖定於影像中心。本文發展的雙目標影像追蹤系統可以同時追蹤二個相同形狀不同顏色或相同顏色不同形狀的移動目標物。 關鍵字:立體視覺、雙目標追蹤、向量支持機器、多重閥值顏色篩選
Detection and tracking of moving targets in live video streaming is one of the necessary functions for an intelligent video surveillance and monitoring system. But if a vision system can be developed to detect and track multiple targets simultaneously, its functions will be more complete. Hence, the purpose of this dissertation is to construct a stereo visual servoing system which can track two moving targets simultaneously. The proposed system consists of two stereo vision platforms, and each platform has two cameras with two rotatory degrees of freedom. Firstly, moving targets in live video streaming can be captured by using image processing and color multithresholding methods. Furthermore, in this dissertation, a support vector machine is built for recognition of moving targets because it has ability to classify nonlinear non-separable types of data. Finally, the centroid of each target can be determined and then the relative positions between targets and cameras can be estimated by using stereo vision algorithm. With converting the relative position into motor control signals, each stereo vision platform can track its moving target and put it on the center of image. The proposed dual-target tracking system can track two moving targets with same shape but different colors or with same color but different shapes simultaneously. Keywords: stereo vision; dual-target tracking; support vector machines; color multithresholding