- 學位論文
利用嵌入式平台實現居家人物觀看數位電視節目之行為分析
Behavior analyses of family members watching digital TV programs via an embedded platform
摘要
隨著科技不斷的進步，智慧家庭已經被視為重點發展項目之一，而電視作為家中客廳之必要家電，其發展的方向更是受到關注，如今隨著有線電視的數位化，家中數位機上盒(Set-Top Box，STB)的存在也隨之普及。在環境與政策的帶動之下，STB產業的競爭也越發激烈，各廠商除了致力於設法降低產品的售價之外，還需要積極地做延伸開發，提供更多元、創新的功能，才能捕獲消費者的心。本論文藉以嵌入式平台實現播放數位電視節目之功能，並擷取人物相關資訊做分析與紀錄，希望能提供STB一種創新的服務與產品開發的方向。 本論文使用揚興科技提供的MPB573 STB開發板做為嵌入式系統的開發平台，藉以實現播放電視節目廣播及人臉辨識的功能。首先針對系統開發套件進行重新組譯的動作，之後將編譯好的系統以NFS(Network FileSystem)的方式掛載至平台上完成系統的建置。平台建置完畢後，首要任務是針對DVB-T訊號的解調函式進行修改，並使用CMake將OpenCV 2.4.9移植到MPB573，得以實現播放台灣無線數位電視頻道與人臉辨識的功能，最終以HTML5提供使用者一套具備遠端遙控的介面及觀看分析結果的功能平台。 在一般情況下，若只使用OpenCV的Face Detection和Eigenface來實現人臉辨識系統，其平均辨識率僅為78.8%。本論文為了能提供更好的人臉辨識系統，因此本論文使用V4L(Video4Linux)取得即時的影像串流做為輸入訊號，同時針對辨識系統加入自適應膚色檢測法與投票機制，經過改良後的平均辨識率提升至85%。
並列摘要
With the progress of technology, smart home has been considered as one of the main developments by IT industry. TV is necessary appliances in living room, and the direction of its development is of concern. Nowadays with cable TV digitization, Set-Top Box (STB) presence also will be popular at home. Under the policy, STB-related industry is getting more and more competitive. Except finding ways to lower the price, the manufacturers also work hard to develop more new function and provide their consumers with more diversified and innovative services in order to win their heart. In this paper, we hope to implement play TV programs based on embedded platform and capture the information of the people who is watching TV. According to these data, we can record and analyze their behavior, further provide a new service based on the STB. In this paper, we use an Evaluation Board of embedded system named MPB573 from Aonvision Technology Corporation to accomplish our goal, broadcasting TV programs and face recognition functions. First, we recompile the system development kit, and use Network File System (NFS) to build the compiled system to MPB573 platform. Second is to modify the DVB-T signal modification function and use CMake to install in MPB573. Last is to provide the user with a remote control interface and a result table function by HTML5. However, if only use Face Detection and Eigenface in OpenCV to accomplish the face recognition function, the recognition accuracy is only 78.8%. According to this paper, with the goal to enhance the accuracy of face recognition, we use Video4Linux (V4lL) to get video streaming and make it as the input signal. At the same time, we add Adaptive Skin Detector (ASD) and voting mechanism for the face recognition system. Finally make the accuracy increase to 85%.