Comparison of video shot boundary detection techniques

This paper is about comparing different techniques used for boundary detection of images by comparing different frames of video.

They state that the easiest way to detect if two frames are significantly different is to count the number of pixels that change in value over a value that is set, however they reported that it was slow.

Shahraray [2] divided the images into 12 regions, and compared the differences.

The paper then goes on to discuss different tests that they preformed, a summary of some of them are below:

1. Used histograms, to detect difference in a 64 bit gray scale frames. “A shot boundary is declared it the histogram difference between consecutive frames exceeds a threshold”.

2. Used regional histograms: each frame is divided into 16 blocks in a 4X4 pattern. The histogram differences are computed for each reagion between consecutive frames. If the number of region differences exceeds the difference threshold is greater than the count threshold then the difference is noted.

3. Motion compensated pixel differences: again the screen it divided but this time into 3X4 , using grey scale to detect difference in the pixels.

All the algorithms were implemented in C and run on Unix.

All input video was digitized at a size of 320x240 pixels at a frame rate of 30 fps using a DEC Alpha equipped with a J300 video board. The digitized video was stored as motion JPEG with a compression ratio of about 25 to 1, requiring about 1 GB of space to store 1 hour of video.

From this paper, the idea of using grey scale seems to be the way it has to be done, and differences in the histogram would be noted and begin the sequence of actions. The area separations would allow for greater control, if for instance we used the dog and we wanted it line of vision to change, e.g. move its head in the direction someone moved. Due to the file size of ongoing frame recording of differences I would think that we would only take into account any 2 frames at one time.

[1]Berkeley Multimedia Research Center
Published: September 1994
Berkeley, CA
USA
http://www.bmrc.berkeley.edu

http://bmrc.berkeley.edu/research/publications/1996/133/shots.html#ab


2. Shahraray, B., "Scene Change Detection and Content-Based Sampling of Video Sequences", in Digital Video Compression: Algorithms and Technologies, Arturo Rodriguez, Robert Safranek, Edward Delp, Editors, Proc. SPIE 2419, February, 1995, pp. 2-13.

Went on to look at Phidgets and at the moment Motion sensors are out of stock, but futher reading showed that some web cams come with motion detection built into them.

http://computer.howstuffworks.com/webcam1.htm