Peanut Butter + Jelly

For computational camera, we were interested in how people shape structures with their movement. Some examples are play-doh, sculpting.

Traditionally this was always done with a motion capture system, which required cameras that costs a fortune and complex setting (such as a blue screen, tracking balls, etc). This is way too expensive. We wanted to see how much information we could get from just a one camera system!!

Ideally we would be able to create landscapes such as the drawing from the great Zaha Hadid.

Posted by ben chao at 12:20 PM | Permalink

Moition History Image

How to produce Motion History Image

1) Convert Image to silhouette
2) Update motion history image
3) Calculate motion gradient of individual elements
4) Find motion orientation of entire image

Motion gesture recognition systems is based on the idea that it is simple for humans to recognize motions patterns even though the image might be blurred.

Example video.

Standard motion signatures or movements are built into a library. By matching motion history viewed on a camera it is possible to recognize certain movements that a person has performed.

Example video.

Applications
- Entertainment System
- Expression Recognition

Motion History vs Optical Flow
- Optical flow requires sharp edges to recognize points of interest, thus requires a high resolution camera and is computationally more expensive. Added benefit is it can recogize general structures which motion history can not.
- Motion history recognizes general "patterns" of movement, thus can be implemented with cheap cameras and lower powered CPUs. It can also be implemented in low light areas where structure can not be easily detected.

Special thanks to Gary Bradski and James Davis for their research paper - Motion segmentation and pose recognition with motion history gradient.

Posted by ben chao at 10:29 PM | Permalink

here's the processing for this week's assignment

Posted by ben chao at 05:18 PM | Permalink