Key Points
We separate body-sway videos into temporal micromovements and the body's spatial appearance. By amplifying the high-frequency components of motion while suppressing image artifacts, we visualize subtle sway that is difficult to see with the naked eye.
Abstract
A standing person's body makes tiny movements that can be difficult to see. We aim to make the faster parts of this sway more visible by magnifying them in video. Existing methods can produce unnatural distortions because they do not reliably separate movement from the body's appearance.
We assume that the small movements are shared across the image regions within the person's head. This lets us separate the video into changes in movement over time and the person's appearance. We then magnify only the faster movement components. These rapidly repeating movements are known as high-frequency components over time.
In our experiments, the method magnified body sway without introducing image artifacts, making the faster components visible. We also applied it to a system that classifies baggage as heavy or light using body sway. The resulting videos helped show why that classification succeeded in some cases and failed in others.
Publication
- 我那覇 航，神谷 卓也, 西山 正志, 岩井 儀雄,
身体動揺に含まれる高周波成分の可視化のための運動拡大を用いた映像生成,
精密工学会誌, vol. 88, no. 12, pp. 926 - 936, December 2022.
Short movies
