Lecturerecorder - Presenter Detection

This video is the output of the detection phase of the http://lecturerecorder.googlecode.com project. We use OpenCV haar detection with an 'upper body' haar classifier. We combine the output with a 'candidate object' algorithm and we use one time room calibration to more accurately detect the speaker.

Black circles are detected as 'upper bodies' but discarded by the candidateobject algorithm. Purple circles are potential presenters. The lighter the purple color, the higher the chance that we are dealing with a presenter. The green circle is deemed to be the presenter by the candidateobject algorithm.

Note: we currently only support a single speaker, which is why the scene where there is a second speaker has some strange but interesting results.

The original input video can be found here:
* http://www.youtube.com/watch?v=dQz_8tOgxj8

The next step (processing the video) can be seen here:
* http://www.youtube.com/watch?v=i2wJwfgrxZc

(there's no sound, we were only dealing with the video bits)

Category:

Education

Tags:

• lecture
• opencv
• computervision
• detection
• motion