觀眾行為感知智慧聯網電視---技術與應用服務

Abstract

聯網電視（Internet TV; IPTV; ITV)已經是目前十分普遍的網路服務，而近年來由於觸控 智慧裝置的迅速發展，各國際大廠紛紛推出各種IPTV或Smart TV之商品，如Goole TV、 Sony ITV、Apple TV、Samsung Smart TV，Window 7 TV 等。這些商品或服務，都不脫離 透過螢幕選單，智慧型遙控與電視機間之視訊串流，甚至可分辨手式以開關電視機等。 本研究計畫中，我們將進一步研究觀眾(收視者)行為感知(Audience-aware)之智慧聯網電 視(Smart ITV)高階技術並發展其應用。首先我們會透過電視機前的感測裝置，收集觀眾 之資料，進行對觀眾的身分及行為模式之辨認，以推測其年紀、習性、或情緒等相關資 訊，作為節目或商品推薦之有利線索。另一方面，利用影像處理的技術，以取得影像中 物件的內容及相關資訊，藉此我們可以找出與影像內容高度相關之類似影像，並分析影 像內容，找出影像的關鍵字以獲取其中重要物件之名稱；接著找出影像的類別以歸類影 像之屬性。透過收視者生理感測資料、與所分析出的電視影像屬性、與使用者的實際觀 看記錄做結合，搭配資料探勘技術，讓智慧聯網電視能夠依據過往的收視習慣，推薦該 收視者可能感興趣的節目與廣告。除此，由於大部份的家庭都擁有電視，智慧聯網電視 將結合電腦功能晉升為家庭的多功能的監控及娛樂中心。透過個人區域網路（PAN)， 可以讓家庭收視者的智慧型手機和電視做資訊交流，並和家中之物聯網元件進行通訊， 監控各種物聯網元件。由於節能減碳的趨勢，本計畫將先著重於耗電偵測，讓智慧電視 藉由PAN傳輸收集各電錶的資料，並將之整理顯示在螢幕上，如此可以輕易知道各種 家電耗能資訊。由於觸控螢幕的漸漸普及，與體態操控的流行，操控的方式越來越直覺， 因此使用手勢做為和電視主要的互動方式是一大趨勢。我們也將研究一套操作手勢，利 用它能夠應付電視的操作。同時為了減輕觀眾操作管理上的困擾，本計畫中將發展一共 同資源管理模組以提供簡易的介面以存取和管理各種資源。同時提供一虛擬化機制，把 各種設備與服務都虛擬成一資源點，將各式實際資源之屬性、資訊和操作方法等封裝於 該點。收視者只需簡單選擇該點即可拖曳分享各種功能與服務。為使各設備與服務之間 有連鎖之互動反應，本計畫將提供一轉換模組以供介接各式不同之資源。透過此模組之 介接與管理模組的連繋，各設備與服務之間，將可以利用一個命令或觸擊(Trigger)，即 達成連鎖之互動反應。透過上述之各種技術與功能模組之整合，本計畫預期發展三套應 用服務：依觀眾行為屬性之內容推播與廣告(Audience-Aware Content Push & Advertisement)，節目之視覺化捜尋(Visual Search of Programs)，與即時互動學習平台服務 (Real-Time Interactive Learning Platform Service)，並完成一雛型系統展示。

Internet TV (IPTV; ITV) is a popular network service nowadays. Due to the quick advances of touch screen technology, major international venders have announced various IPTV and Smart TV products, such as Google TV, Sony ITV, Apple TV, Samsung Smart TV, and Window 7 TV. These products or services, all feature touch screen, icon menu selection, smart remote control and streaming between handset and TV, few of them feature gesture recognition to turn on/off the TV. In this project, we will make study on advanced-techniques of audience-aware smart ITV as well as develop its application services. First, we will collect the audience information through the sensing devices attached to the TV, then recognize the audience’s identity and his/her behavior model, so as to estimate the information regarding audience’s age, habit, or mood, which can be the necessary information for recommending programs or merchandise. On the other hand, we could use image processing techniques to retrieve the object and related information in the image, so we are able to find the similar images which is highly related to the target image, and analyze the image content to get the key words regarding the name of the important object. Then we could classify the image into category of properties. Through the information of physiological sensing and the analyzed property of TV images, the audience’s watching records, as well as data mining technique, smart ITV should be able to recommend either programs or advertisements that the audience may feel interest based on his/her past watching habit. Besides, since most homes own TV, smart ITV can integrate TV and computers to become a multi-purpose monitoring and entertainment center. Through personal area network (PAN), it allows home audience’s smart handset to do information exchange with TV, and communicate with devices connected via Internet of Things (IoT), so as to monitor and control all IoT devices. Due to the trend of energy saving and carbon reduction, we will emphasize power consumption detection first, and allow smarty TV to collect data from meters through PAN, then process the collected data and show it on TV screen. The audience can easily obtain the energy consumption information of his/her home. Because the touch screen has been widespread, the physiological control is popular, and the control manipulation becomes intuitive, therefore interaction with TV through gestures also becomes a trend. We will investigate a set of gestures, through which we could handle the control operations of TV. In the mean time, to reduce the difficulties of operation management, we will develop a common resource management module to support a simple and friendly interface for access and manage various resources. The module will provide a virtual mechanism that virtualizes all devices and services into a resource point, and encapsulate the property, information, and operation method into that point. An audience only needs to simply select that point then he/she could drag and share various functions and services. To enable chain interactions between various devices and services, we will also provide a transformation module for interfacing between various resources. Through the interfacing of transformation module and interconnection of management module, chain interactions will occur between various devices and services with either a single command or a single trigger. Through the integration of techniques and function modules described above, we expect to develop three application services: Audience-Aware Content Push & Advertisement, Visual Search of Programs, and Real-Time Interactive Learning Platform Service, and complete a prototype system for demonstration.