方瓊瑤Chiung-Yao Fang蔡侑廷You-Ting Tsai2019-09-052015-6-282019-09-052013http://etds.lib.ntnu.edu.tw/cgi-bin/gs32/gsweb.cgi?o=dstdcdr&s=id=%22GN060047037S%22.&%22.id.&http://rportal.lib.ntnu.edu.tw:80/handle/20.500.12235/106551本研究的目的乃在於建立以觀眾為拍攝主體之虛擬攝影師系統。由於人力成本的不斷提高，聘請商業攝影團隊為演講或會議拍攝影片對一般預算有限的非營利學校或公司行號是一個沉重的負擔。若為了節省成本而讓缺乏攝影經驗的人來掌鏡，往往會讓影片缺乏美感及流暢度，甚至是降低觀賞的意願。本研究所提出之系統除了能夠節省人力開銷，還提供專業攝影技巧以製作高規格影片。 本研究以兩台PTZ攝影機作為實驗設備，架設於拍攝場地的左或右前方，兩台PTZ攝影機具有不同的功能，一台為全景PTZ攝影機，另一台為運鏡PTZ攝影機。全景PTZ攝影機代表攝影師的雙眼，將拍攝場地的全景的連續影像輸入到系統協助系統的全景監控與主體偵測。運鏡PTZ攝影機則是攝影師手上的攝影機，在系統決定運鏡所需之一切資訊後，運鏡PTZ攝影機就會實際執行運鏡動作。 本系統的功能為模仿攝影師的拍攝技巧並自動進行運鏡動作，每次運鏡皆需要運鏡的種類、景別、主體等要素。系統從全景PTZ攝影機輸入連續影像，從中擷取四種具描述觀眾行為的motion特徵。接著將這些特徵經過模糊化(fuzzifierion)處理讓這些數值轉換成自然語言的表達方式，以便於分析攝影師的運鏡習慣。接著將這些特徵輸入自動運鏡模型(automatic camera movement model)，該模型能夠紀錄專業攝影師的運鏡習慣，並依據輸入的特徵輸出適合該情況的運鏡以及景別種類。拍攝主體的挑選則以五種分別代表不同美感的特徵作為判斷標準。在完成運鏡要素的計算後，將資料傳送至運鏡PTZ攝影機執行運鏡動作。 實驗結果顯示，本系統可進行即時且流暢的運鏡動作且具專業攝影的運鏡水準，符合演講錄製的需求。This study proposed an automatic virtual cameraman system for audience. Because of increasing of personnel costs, to employ commercial photography team filming speeches or conferences is a heavy burden for the nonprofit school or company line number that budget limited. If In order to save costs and let person who lack of photography to shoot, it often causes the videos which not smooth and aesthetics. Our system not only can save personnel costs, but makes professional videos with photographic. System equipment includes two PTZ web-cameras which are located right front or left front of the lecture theatre. These two PTZ web-cameras have different functions. One, which named full-shot PTZ camera, grabs auditorium screen and input continuous images to system. Another one , which named camera movement PTZ camera, do camera movement after system computes all necessary camera movement information. Camera movement information includes camera movement class, shot class, and subject. To get this information, system first input continuous images from full-shot PTZ camera. Then extracting four motion features which represent four kinds of audiences’ behavior. For combining motion features and natural language, system fuzzifier motion features. To decision camera movement and shot, we construct automatic camera movement model (ACMM), an automata model. ACMM records photographers’ habit of camera movement and shot. It can pick up suitable camera movement and shot by input fuzzy motion features. After that, system will choose subject by using four aesthetic features, which are continuous, repeated, luminance and composition. Last, system operates camera movement PTZ camera to finish recording. In the experiment, there are eleven kinds of camera movement and six kinds of shot in the ACMM. And ACMM was trained with six lecture videos. Compared to amateur’s recording and point shooting, our system is really performance well than them. Beside, system can real time work, so it won’t lose any shot on the auditorium.攝影師運鏡美學自動機cameramancamera movementaestheticsautomata