Publication,
Research,
Docs and Info,
KAMEDA
We have proposed a human motion estimation method from three successive video frames. As it is no need to estimate a motion of a person when he stays still, we propose to use a double-difference image to get motion information from video frames and estimate his motion only if the motion regions exists on the double-difference images. We implemented a motion estimation system that executes model matching on a double-difference image and showed its abilitiy.