Abstract:

In the proposed paper we describe an Automatic Video Editing System designed to record and edit classroom and conference videos. The system supports different video and audio sources (teacher´s and students’ camera/microphone, slices presentation, screen capture, PC audio, Digital Blackboard…) and automatically recognizes the key events of each video and audio sources in order to mount an automatically generated intelligent cut. The richness of different types of video sources and events that it can manage allows this system to recognize and to deal with a wide variety of different session profiles (master class based on traditional/digital blackboard, slices presentation, discussions, questions and answers, software and devices handling and many others) and provides a base software to expand the system to videoconference based class sessions.
This system will allow students to review class sessions for a better understanding of the contents, it will let ill students to follow the classes from home, avoiding in this way, losing their classmates beat and this system will also give the possibility of e-learning for remote courses and conferences. Furthermore it could be used by teachers in order to analyze their verbal and non-verbal communication capabilities like speech tempo and emphasis, explanation´s clarity, exposition´s appearance, etc. or it could be used to observe the students´ behavior during the session so they could, afterwards, improve their communication capabilities and educational effectiveness.
We have tested the system on a master class session based on blackboard and slices presentation, permitting real time questions and final discussions between teacher and students. For this kind of scenario we have used 2 fixed cameras, one of them framing the teacher, blackboard and slices projection screen and the other one framing the students, a screen capture of the slices presentation and a wide variety of microphones (teacher´s lavalier microphone, teacher-oriented ambience microphone, students-oriented ambience microphone and a wireless hand microphone for the students). Then we have analyzed the behavior of the event segmentation system with each audio and video source, getting the best event recognition results using the slices presentation (slice change event), the lavalier microphone (teacher´s speech event) and the wireless hand microphone (students´ speech event), but also getting good results using the 2 ambience microphones, which seems to be much more realistic, cheaper and an easier way to do it. Furthermore we have included sporadic controlled shot changes in the algorithm to avoid the large and boring shots in the cut as well and to get a more natural edition.
After the recording, the event automatic recognition and the automatic editing cut, we have sent to students and teachers a portion of the test automatically edited video and a survey. We can conclude a good acceptance of the system based on surveys´ results (up to 70% of 67 teachers and 95% of 70 students), most of them say will use the system but many of them will make it conditional on their use control. Both students and teachers think it will help them to improve their teaching-learning processes and the most desired applications are for conferences and hardware or software handling classes followed by theory classes. Finally, the subjective perception of the cut quality is good for most aspects, while a few require further work.

Keywords:

Automatic video editing, automatic event recognition.