3rd year projects 2005 - 06

  • JPB-UG-1: Video Story-Board Extractor and Player (Nick Firth)
  • JPB-UG-2: Blink Rate Estimation (Andrew Savill)
  • JPB-UG-3: Video Pan and Zoom Analysis (Daniel Melia)

Mail all

The project descriptions below are only intended as starting points. If you wish to discuss possibilities in greater detail I encourage you to email me to arrange a meeting.

JPB-UG-1: Video Story-Board Extractor and Player

Description

Video takes up an enormous amount of storage. Compression schemes such as MPEG reduce the amount of storage required by exploiting the similarity between consecutive frames. However, video could be compressed much further if you are not concerned about preserving the illusion of motion, but only about conveying the ‘meaning’ of the video. At the extreme the video could be reduced down to a series of still `key’ frames, similar to the story-boards that film makers use before shooting a film.

The aim of this project is to produce a tool that can automatically reduce a television program or a feature film to a minimal sequence of still images required to retain the ‘meaning’ of the film. One approach would be to select a frame that is representative of each `shot’ of which the film is composed. A second tool will enable the sequence of still frames to be replayed in synchrony with the audio track from the original video.

The idea for this project was inspired by the French film, ‘La Jetee’ (http://www.imdb.com/title/tt0056119/) which is entirely composed of a sequence of static shots (and which was also the inspiration for Terry Gilliam’s ‘The Twelve Monkeys’).

Software may be developed either in C/C++ using a computer vision library such as OpenCV, or in Java using the Java Media Framework.

Requirements

  • good java or C/C++ programming skills</BLOCKQUOTE>

Reading

[TOP]

JPB-UG-2: Blink Rate Estimation

Description

There is currently much interest in developing techniques for judging Human emotion. Being able to recognise emotional state is crucial to the development of natural Human-Computer communication. There are many clues that Humans use when judging emotional state (e.g. tone of voice, appearance of the face). One of the more subtle cues is the rate at which people blink. Blink rate also an indicator of alertness, and blink rate estimation has application as a means of monitoring driver alertness for car safety.

This project will attempt to use some straightforward video processing techniques to construct a robust `blink detector’. A blink is readily detectable because it involves rapid movement occurring in two different places (i.e. both eyes) at the same time. By looking at the pattern of differences between subsequent video frames and having a crude model for the geometry of a face it should be possible to detect such an event with a high degree of reliability.

The project will employ an audio-visual corpus recently recorded at Sheffield, in which 34 people have been recorded speaking to a camera. Software will be developed either in C/C++ using a computer vision library such as OpenCV, or in Java using the Java Media Framework. The project will be evaluated by comparing the results of the system with hand-annotated video.

Requirements

  • good java or C/C++ programming skills</BLOCKQUOTE>

Reading

[TOP]

JPB-UG-3: Video Pan and Zoom Analysis

Description

All film directors have their own style. One element of this style is in how they use the camera. Some directors prefer to keep the camera movement as minimal as possible, others make greater use of camera pan and zoom.

This project will use ‘optical flow’ techniques to describe film shots in terms of camera motion. By analysing longer sequences of shots it should be possible to build a crude description of the style of a piece of film. Such a description could be used for video classification, identification and retrieval applications.

Software may be developed either in C/C++ using a computer vision library such as OpenCV, or in Java using the Java Media Framework.

Requirements

  • Good Java or C/C++ programming skills</BLOCKQUOTE>

Reading

[TOP]