Aqua Phoenix
     >>  Research >>  Lecture Video Browser  


4. Observations and Experimental Data

Depending on the frequency and type of usage of a media type, segmentation can be very subjective. Below are a few guidelines that arose from observation.

4.1 Usage of the Board

Should the instructor make extensive use of the board, it is useful to cluster Slides by their written contents into Topics. We define a Topic as Slides with same content; however, this is an arbitrary definition and can be challenged with other interpretations. In our definition of a Topic, the written/illustrative material across Slides can be tracked by matching (superimposing) the content. Most of the time, the content of a Topic appears in a well-defined physical area of the blackboard (or whiteboard), such as physically beveled board panels.

Keep in mind that blackboards and whitebords differ greatly in design - some are entirely fixed without moving components, some contain folding panels, some have the ability to slide vertically, yet others have two or more sliding parts. Depending on the design and instructor's use of the board, a Topic can be manifested in several ways. In our analysis of lecture video, we have found that a Topic is best defined as matching content within physically bounded areas, i.e. blackboard panels. The camera operator tends to use the confines of these panels as fields of view for the recorded material as well.

Should the instructor use one particular board panel filled with content throughout the entire lecture while erasing and replacing only small parts, the paradigm of location-based Topics shifts. In this case, a change in Topic may be better defined by the moments in time at which parts of contents are erased. This paradigm stands in contrast to our observations, in which either the entire board panel is erased or a separate fresh board panel is used for a new Topic.

4.2 Usage of Sheet (projected material)

We find that instructors use sheets of paper in conjunction with projection as an alternative to the blackboard. The two generally follow the same characteristics with few subtle differences: 1. Written content on sheets are not erased at any point, 2. The physical boundary of a sheet of paper tends to dictate a Topic. Therefore, for media type Sheet, Topics eventually refer to single sheets of paper.

4.3 Slides from the Class

We found that this media type appears very rarely. Especially in a engineering lecture videos, which are used for distance learning and expert instruction, liberal arts-style discussions are rare. If questions by students do occur, the camera operator may decide that it is not worthwhile shifting the camera to the student, because the audio track captures the necessary information.

Nonetheless, if Slides of media type "Class" do occur, the structured lecture content does benefit from their separate classification. In a visual interface, these Slides can be easily identified for "question" or "discussion" content.

4.4 Usage of the Computer

At the time of this research, few instructors made extensive use of the computer. Electronic slides (PowerPoint, etc.) were used rarely and only experimentally.

Electronic slides cannot be tracked in the same fashion as blackboard writing: contents across slides, while contextually linked, changes instantly (i.e. change of slide), and therefore, an optical flow of written material cannot be clearly established. Reasonably well-defined Topics may need to be established through manual analysis, i.e. human understanding of the contents.

When the instructor has structured the electronic slides carefully into sections, e.g. by providing a table of contents or numbering titles by sections and sub-sections, this information can be applied to Topic identification.

From the courses we have analyzed, electronic slides appears rarely and often in sequential sets of two to five. The instructors used the electronic slides to quickly summarize the discussed material and to introduce new material. We did not find it necessary to further segment these slides into separate Topics - consecutive electronic slides are therefore clustered into one Topic.

4.5 Slides from the Podium

The occurence of Slides of media type Podium depends highly on the camera operator. Podium Slides are defined as the instructor addressing the students without the immediate use of material (board, paper, computer, etc.). This usually occurs during course announcements (e.g. class assignments) and discussions, in which the camera shows a far view of the podium and the instructor. However, even during a session in which the blackboard or other material are used, the instructor frequently addresses the class without explicitely writing.

For example, after writing material on the board (or a sheet of paper), the instructor turns to the class to elaborate, before continuing to develop the written content. At this point, the camera operator must make an educated decision as to the camera's field of view in order to best capture the flow of the course: 1. keep focus on the blackboard where content has just appeared, or 2. zoom out and capture the instructor's address to the students. A well-trained camera operator knows the difference by following the course attentively. Typically, during short addresses, the camera remains focused on the content, while during longer addresses, the field of view changes. This best reflects the flow of the lecture and the intent of the instructor.

The effect of over-sensitivity to changes in the instructor's focus, i.e. focus on blackboard versus addressing the students, is nonetheless mitigated by our design of Topics. The frequency at which the camera operator changes the camera's field of view between content and the instructor addressing the students is eventually reflected in the visualization. A set of Slides with long periods of time spent on content and few breaks showing the instructor addressing the students results in a compact representation containing long sequences of Slides per Section (Section = temporally consecutive Slides in a Topic). On the other hand, when the camera view switches frequently between a Topic and a Podium shot, the representation of the contents will be much more fragmented with many short sequences of Slides per Section. Nevertheless, the visualization still shows the progression and coherence of the fragmented Sections per Topic.

It is possible that a professor presents most of the class without visual material. In this case, good judgment or good speech recognition and analysis is necessary to generate Topic segmentation.

By convention, the title slide ( title.jpg or 0.jpg ) is a Podium slide, as Podium slides tend to appear in the beginning of the class, during which the class name is announced.