以視覺為基礎之師生互動行為分析系統

Vision-based Analysis System of Interactions between Teacher and Students

林子怡(ZI-YI LIM)

指導教授 ： 廖珗洲

朝陽科技大學/資訊學院/資訊工程系/碩士(2016年)

摘要

課程錄影系統主要作為錄製學生課後複習的影音教材，而本研究的動機來自於一個產學合作案，公司期望在多攝影機的環境中可以錄製教師與學生在課堂上的互動過程，並自動將多個影片自動剪輯為單一影片。本研究中，教師端使用1台廣角攝影機，學生端則依據教室大小使用2台以上的攝影機進行分析。依據分析教師與學生互動所產生的事件，將會在相關攝影機的影片上記錄標記(Tag)，以利於公司現有的課錄系統可以在課程結束後依據標記進行自動剪輯。
教師端的分析上會先在攝影機的FOV (Field-of-View)預先設定一組區域，而教師端分析的主要目的是可以確認教師的所在區域。分析方法使用GMM (Gaussian Mixture Model)建立背景模型，再將即時影像進行背景相減取得前景物體，並將最大的物體將視為教師，在教師的座標上也使用了卡爾曼濾波器來避免座標跳動，最後再利用一組條件過濾的機制確認教師所在區域。
學生端分析的主要目的是為了取得學生站立的事件。分析方法同樣使用GMM來取得前景物體影像。接著過濾物體的幾何特徵以確認是否為學生，並進一步分析區塊移動方向以確認學生是否有站立行為。
上述分析所得到之教師或學生的分析事件透過HTTP的方式傳送至課錄系統，經過公司實際與安裝測試，本研究所開發之系統已完成商品化，並且公司將系統命名為“iTrace”實際列入公司的服務項目中。

關鍵字

視覺處理 ； 教室錄影 ； 教師行為分析 ； 學生行為分析

並列摘要

Lecture recording is an important function for students to review the lesson after class. In an academic-industrial project, the interactions between the teacher and students are expected to be recorded in a multiple camera environment. A wide-angle camera is installed for the teacher and two or more cameras are installed for the students according to the size of the classroom. The interactions are then analyzed and generate events to put tags in the corresponding video description. Then, all the videos recorded from different cameras can be blended based on the tags into a single video after the class automatically.
A set of areas is pre-defined in the FOV (Field-of-View) of the teacher’s camera. The purpose of the teacher’s video analysis is to get the correct area where the teacher is contained. In the analysis of the teacher events, GMM (Gaussian Mixture Model) is used to construct the background model and segment foreground objects. The largest blob is deemed as the teacher. The Kalman filter is utilized to keep the teacher’s coordinates stable. Finally, a set of rules is designed to ensure the area is correct.
The purpose of the student’s video analysis is to get the stand-up events. The same foreground object segmentation method of teacher’s analysis is used. The geometric features are filtered to ensure the blob is a student. Then, the moving direction is also checked to ensure whether the student stands up or not.
The above teacher or student events are transmitted to the lecture recording system via the HTTP connection. The system is commercialized and becomes a service item of the company called “iTrace”.

並列關鍵字

Visual processing ； Classroom recording ； Teacher’s behavior analysis ； Students’ behavior analysis

參考文獻

[7]H. C. Liao, and M. H. Pan, “An Automatic Lecture Recording System Using Pan-Tilt-Zoom Camera to Track Lecturer and Handwritten Data,” International Journal of Applied Science and Engineering, Vol. 13, No. 1, pp. 1-18, 2015

[1]M. Bianchi, "Automatic Video Production of Lectures Using an Intelligent and Aware Environment," Proceedings of the 3rd international conference on Mobile and ubiquitous multimedia, pp. 117-123, 2004.

[2]F. Lampi, S. Kopf, M. Benz, and W. Effelsberg, "A Virtual Camera Team for Lecture Recording," IEEE MultiMedia, Vol. 15, pp. 58-62, 2008.

[6]M. B. Winkler, K. M. Hover, A. Hadjakos, and M. Muhlhauser, "Automatic Camera Control for Tracking a Presenter during a Talk," IEEE International Symposium on Multimedia, pp. 471-476, 2012.

[9]D. Hulens, T. Goedeme, and T. Rumes, " Autonomous Lecture Recording with a PTZ Camera While Complying with Cinematographic Rules," Canadian Conference on Computer and Robot Vision (Crv), pp. 371-377, 2014.