Shape From Focus (SFF) has been known as a depth estimation technique by using focus information. The technique estimates the depth for each point of a target object based on a focus measure extracted from a set of multi-focus images. The authors have extended the technique to a multiple-viewpoint algorithm. However, the technique has been leaving the problems that the number of viewpoints is limited to 2 and that large reconstruction error occur due to a lack of visibility checking from each viewpoint. This paper thus tries to increase the number of the viewpoints and reduce the error by introducing a hidden surface detection.
This report proposes an interactive system for public space called "ResponsiveWall", which senses ambient conditions around it and makes graphic representation in response so that people around it spontaneously arouse their interests and interactions. The system is composed of several chained units, each have a PC with a large touch-sensor equipped display and a camera with a mike on top of it. The processing is performed using a number of simple modules each responsible for processing a part of sensing information and make responsive representation on a part of screen. The system realizes dynamic and natural interactions between it and people in public space.
This report proposes a system called "MonoReco" which allows us to record a voice message as if we put it into a real object and replay it. We aim to enable people to interact with real objects unconscious of computers behind the scene. This is made possible by just using a scene image to link recorded voice to a user-pointed object. Voice recording and replaying are triggered by detecting a colored card shown by the user. It is very easy for users to use the system by just showing a colored card over the target object and saying something in recording and listening in replaying.
We have proposed an extraction and tracking technique for image sequence of living cells, and realized a cell image analysis system. However, the efficiency and accuracy of the technique drastically vary depending of the image acquisition conditions. To alleviate the problem, this paper introduces all-in-focus images and automatic threshold determination. The experimental results given in this paper illustrate that the proposed technique improves the efficiency and accuracy in the extraction and tracking process over the previous technique.
We describe our project to establish collaborative safety mechanism based on mutual sharing of driver's information. Most conventional driving safety approaches aims at assisting individual drivers by providing various safety information to the driver on the vehicle. However, the effect of such system highly depends on drivers' response. In our approach, we try to make the driver's present performance to be shared with other drivers and pedestrians. Using this sharing mechanism, people in traffic will be able to make better prediction of other driver's behavior and it leads to their proactive actions to prevent accidents.