自動化演講錄製系統

Abstract

在本論文中，自動畫演講錄製系統，期望能以導演運鏡的方式記錄下整個演講內容。整個系統由兩台PTZ攝像機構成。一台負責追蹤講者，另一台負責拍攝投影布幕畫面。首先，利用Adaboost分類器找出講者臉部所在位置，接著以Mean-shift演算法對講者臉部區域進行追蹤。如果講者使用教具輔助演講，如雷射筆，指揮棒，和手勢，可以偵測出其位置並標示出來。與此同時，另一台拍攝投影布幕的攝影機利用強度閾值，定出投影布幕區塊進行拍攝。基於講者和投影布幕的相關位置，系統自動調整講者的臉部大小，方向，位置，使其符合專業攝影師所預定義的審美標準。

In this paper, an automatic lecture recording system, which is expected to be able to act as a professional photographer, is developed. The current system consists of two PTZ cameras. One is for tracking the speaker and another one is for localizing the screen. To begin, the system locates the speaker’s face in the input image provided by the first camera using an Adaboost detector. The located speaker’s face is then tracked over the video sequence using a mean shift algorithm. If the speaker uses teaching aids such as laser pens, batons, and hand gestures, they can be detected and marked as well. In the meantime, the screen is targeted at by the second camera simply by intensity threshold and then tracked over the video sequence according to the movements of the camera. Based on the information about the speaker and the screen, the system automatically adjusts the speaker’s size, orientation, as well as position to be present in images according to some predefined esthetic criteria suggested by professional photographers.