Firstly, how to deal with privacy in videos? Secondly, since there is a lot of video content to watch on the web - how to keep an overview and use the time effectively?
We applied recent research results: To preserve privacy, we detect objects - especially faces - and apply intelligent stylization filters ("cartoonify") to parts of the image that contain the face. For video summarization, we run a shot detection and selection of most important frames of each scene.
I worked on the pipeline architecture of the whole project. Also, I worked on the face detection part of the pipeline. I used LBP & Haar classifiers, eye classifieres, and classifiers for profile detection. Tracking of detection frames was done with meanShift and camShift.
FFmpeg, OpenCV, OpenGL, C++