Slides from Video Lecture

The project main goal was to design an algorithm in order to create slides from a lecture film; these slides should contain all the relevant writing from the board.

Abstract
The project main goal was to design an algorithm in order to create slides from a lecture film; these slides should contain all the relevant writing from the board. Hence, the input of the project is a video film (Or a sequence of pictures generated from the film) and the output is a collection of slides with the relevant writing. In order to implement that, a program should track the film and collect these slides.

The problem
Two main problems stood before us. One was to detect where the lecturer in the picture is. The second was to collect the relevant data from the movie in order to create a useful lecture slides “book” that contains all the writing in the lecture.

The solution
1. Detect the lecturer’s figure
In general two ways were suggested for the first:

a. Separate the lecturer from the picture by color space (HSV,RGB or other)
b. Separate the lecturer from the picture by edge detection

2. Collect the writings data

The algorithm
1

Results
2

Tools
1. Matlab program (Written on 6.5 version) on any standard P.C. from the VISL lab
2. Adobe premier was used to convert the film to BMP’s
3. CD’s

Conclusions

  • The main problem encountered was the repetition of slides
  • Thanks to the program generality, the program can be used for any purpose that needs to picture a fixed image hidden behind moving objects
  • A better mask can improve the program performances
  • The program could be ran directly on the film, with minor additions to it

Acknowledgment
We want to thank all these who helped us during the course of the project:
Guy Gilboa, Johanan Erez and Ina Krinski.
We are also grateful to the Ollendorff Minerva Center Fund for supporting this project.