Publication,
Research,
Docs and Info,
KAMEDA
Automatic input and analysis of human motion have attracted attention from virtual reality and human interface researchers. The analysis by sensors or special devices were implemented and users find it useful these days. However, such special devices tend to impose a burden to measured persons and are not common in our daily life.
Computer Vision, especially using one camera, becomes main technical trend for human motion estimation and many researchers have contributed to it [5], [6], [7], [], [4]. We have proposed to estimate human motion based on precise human shape model where input video frames are binarized in advance [1], [2].
On estimating a human motion by observing video frames, it is important what kind of features to be extracted from the video frames. We have concentrated on silhouette images in our previous research [2], but it is not easy to obtain the silhouette when the background or light conditions changes. In this paper, we propose to use difference images, because the difference operation derives motion information of the object in the images. If the human body does not move the difference value becomes zero, which means it is not necessary to estimate its motion.
Our method introduces a model matching method. A human body model consists of several solid objects each of which represents a part of a human body and has a joint attribute. Here, motion estimation is defined to estimate the value of the joint angles in the model.
We conducted the experiments on the real video frames. The results show that our method can keep tracking the motion of the human body.