It is important for systems to recognize user actions using sensors and cameras when constructing interactive systems and artworks. Conventional systems have tackled to make systems recognize wide varieties of user actions and to install sensing devices with various environmental restrictions. However, since the recognition methods in conventional systems are specialized to their own work, they cannot be applied to other systems and a specialist for activity recognition is required to construct the systems. In addition, conventional systems took a long time to select recognition algorithms and to set the recognition parameters. This means that they cannot have enough flexibility to change the actions to be recognized or to adapt to changing environments. This paper proposes a method of adding interactivity to various surfaces and recognizing the positions and intensities of performed action by using multiple accelerometers. Our method has functions that enable easy settings and maintenance even by beginners in activity recognition. Participants in an experiment on constructing interactive surfaces constructed a system that could recognize two actions at two points in 51 minutes on average. Moreover, we confirmed the effectiveness of our approach with two actual artworks in long-term media-arts exhibitions.

View full abstract

We address the visualization problem for analyzing collective human behavior using Social Media data. Recently, attention has been devoted to constructing sightseeing guide systems that exploit the information revealed by the collective behavior of users in photo-sharing websites such as Flickr and Facebook, where photos are annotated with GPS locations, time-stamps, photographers, etc. Previous work discovered popular photo spots from a large number of geo-tagged photos, and visualized them on maps. For each popular photo spot, we focus on its burst season as a candidate for its attractive period. For instance, Kiyomizu which is a main sightseeing spot of Kyoto in Japan has a number of the burst seasons that attract many visitors in the season of cherry blossoms or autumn leaves. By effectively visualizing burst seasons, we aim to increase the sophistication of a sightseeing guide system based on collective wisdom. We quantitate degree of the attractiveness and propose a visualization method that visualize the collective behavior and spatio-temporal information all at once, and can effectively analyze and compare the popular photo spots in terms of burstiness. Using Japanese Flickr data, we demonstrate that the proposed visual analysis method is effective.

View full abstract