Semi-automatic and easy creation of learning friendly OCW video content
OCWC Global meeting 2011Semi-automatic and easy creation oflearning friendly OCW video content CSatoshi SHIMADA1, TadashiNAKANISHI1, Akira KOJIMA1and Y hi i FUKUHARA d Yoshimi 2 1NTT Cyber Solutions Laboratories 2Meiji University y j y
MotivationCreating a learning-friendly edited video, for video sharing g g y , gon the web, is painful and time-consuming. Take time Good skills required Costs moneyHow to make this process easy?At the same time how to make edited video learning time,friendly ?The goal is to automatically extract information from alecture video and use them to produce a better version ofthe video.
ProblemWe need to take a video and edit it.The cost of video capture can be reduced by fix HD camera.Wide angle HD camera can record whole of lecture. However, HD video is not suitable for sharing on the Web. , gBecause, (1)File size of HD video is too large (2)Slide in simple downsizing video is i unreadable d bl (3)The entire scene is boring ( ) Video captured by fixed cameraVideo editing manually takes much time time.
Proposed method Proposed methodProposed workflow for the easy and better creation Cature the entire scene to a video birds-eye view by Full HD camera Video file is moved to Laptop Automatic detection of important Information Semi-automatic S Screen/Speaker/Slide change /Sp a /S a g transformation Post-production Post production (Mixing) Screen + Speaker + Slide change information produced video
Principle 1920 pixels i l- speaker- Screen or black board- atmosphere of hallare assembled based on 1080 pixelsediting template-projection di t ti of screen j ti distortion f-brightness of speaker brightness Editing template A Editing template B Editi t l t are corrected corrected. 720 pixels 400 pixels i l
Video processing Video processing(1)Speaker detection find faces in the image detection: linear Interpolation is conducted for no face image(2) SScreen d t ti detection: fi d a rectangle under appropriate find t l d i t conditions, size and vertex angle diti i d t l Because th camera can b positioned almost everywhere B the be iti d l t h in the lecture room a homography is used to map the room, screen coordinates to a flat rectangle as if it was seen from the front of it it.(3) Chaptering: Based on a subtraction method upon N equidistant frames to determine if a ‘Slide change’ has occurred or not in ‘Screen region’ S
Usability evaluation15 Participants without prior knowledge were asked howthey reviewed the p y post-processed video in comparison p pawith the original one. gVideos used: 4(=20min×2, 1hour×2) ,720×408 pixels Comparison of M C i f Mean O i i S Opinion Score Original video Proposed method Original video(simple down conversion) S Screen visibility y Speaker visibility Not boring g Presence Overall impressionPostP t processed video d id for learning use 1 2 3 4 5by the proposed method bad good
Other example Other example Original id O i i l videoFacial search area Screen area set by manually Post processed video by the proposed method at the b i of l h begin f lecture at the middle of l h iddl f lecture
Video sharing on the Web Video sharing on the Web Enhance the video sharing function by using SceneKnowledge, which provides a user-friendly web interface to view videos, annotate them and post comment. Post-produced P t d d lecture video AnnotationChapter is displayed within a certain chapter of theComment input form video SceneKnowledge: A Video scene-based video sharing and comment posting system
How to use our software How to use our softwareSceneEditor is a client software implementing the proposed method.Minimum operation is as follows1.1 determine the editing duration2. Set facial search area3. Set clipping area of speaker4. detect screen a a . area5. select or set an editing template g p6. comand video procressing --- auto processing ---Time required for automated post-production is; post production Roughly correspond to the time of the video itself