Publication,
Research,
Docs and Info,
KAMEDA
We have installed this prototype system in several places and have operated the system.
Table 3 shows the number of generated videos for the past three years and Figure 8 shows an example of generated video. We set two systems, one is at the entrance of our laboratory and the other is on the wall inside our laboratory. The videos were generated for every 30 minutes starting at each o'clock in 1997 and 1998, and every 60 minutes starting at each o'clock in 1999. The authors put the results on WWW[2].
Figure 8: Example of Skimmed Video
With this experiment data, let us show the skimming rate of videos by taking up 495 videos stored in September 1999. The camera shot the entrance of our laboratory and each video was taken for 1 hour in which the program processed 35549.4 frames on the average. This means the current system processed 9.875 frames/second at image size of 320 pixel by 240 pixels. The average number of stored frames was 92.37 per video, that is only 0.26 % of the whole video sequence.
When a camera is set in a building, most popular events that can be imaged with static camera environment is passing of a person in front of the camera. The videos taken within the experiment include a lot of passing persons.
For example, results taken on September 20th in 1999 are shown in Figure 9. Number of persons who passed in front of the camera is also shown on Figure 10. From this result, one person took 11.61 frames to pass in average, that means 1.18 seconds.
Figure 9: Stored frames on Sep.20, 1999
Figure 10: Passing Persons on Sep.20, 1999
