Professor Yagi has been working on research to develop "Computervision", a computer processing system with visual sensory capabilities similar to humans. The technology related to this research has its roots in machines used to read postal codes on envelopes, and nowadays in more familiar ways in a wide array of fields, such as face recognition functions in digital cameras, fingerprint readers, and automobile collision prevention systems. Professor Yagi's research team is working to address the fundamental issues surrounding Computervision with a variety of applications in mind.
An approach to “seeing”
I am currently carrying out research in four themes. The first is geometric analysis that measures shapes in three dimensions. For example, I am researching methods for projecting a special light pattern onto an object to measure that object’s depth in real-time, and then recreate that object based on those measurements. The second theme is research in optical analysis that studies the reflective properties of surfaces and the scattering characteristics of light refracted within objects. My recent work in this research includes imaging techniques for visualizing the insides of living things, and the analysis of the reflectivity of different materials.
The third theme is “sensing”. Insects, birds, horses, and humans all possess optimized senses of vision. A computer surely can also have a sense of vision optimized for its purpose. For almost 25 years, my research lab has worked on an “omni-directional camera” that can take a single 360-degree photograph.
The keyword is “people”
Wide-area surveillance imaging captures individuals at a resolution of 30 pixels, which is nowhere near enough to be able to distinguish features such as faces. How can such a limited amount of information be used for identifying individuals? I think many people have experienced seeing someone from a distance, and while they could not see that person’s face clearly, the person’s body shape and manner of walking were clues that revealed that person’s identity. This is a direct example of how people can distinguish among individuals through their movements, such as walking. Our fourth theme for our research is for technology that can distinguish individuals based on the way they walk. This research has been gaining attention as it is the only biometric method of distinguishing individuals in wide-area surveillance imaging. In 2009, our technology was actually applied by Japan’s police force in the first scientific operation of its kind, and it contributed to the capture of criminals.
Research begins with a dream
The experimental work in our lab involves a lot of time and effort put into theory, programming, systems development, experimentation, and research in a process that many would consider tiring or boring. However, I really like working on “Computervision”, and I find the energy to continue because I have a dream. The most important thing I ask of students is that they too have a dream, even if it is not yet clearly defined. In any case, I want you all to have your own dream, and to carry it forward studying in the Humanware Innovation Program. I have great expectations for all of you.