When watching the motion of a thrown ball, humans can predict where that ball is likely to land. However, cameras lack the visual perception to predict the future. A video that has been recorded by a camera is like a flip book made using multiple photographs.

I am researching methods to enable computers to grasp motion in a manner akin to the human eye. This field is known as “computer vision” or “robot vision”—and I specialize in “motion estimation,” which seeks to determine the motion of various subjects captured in a video.

Research into motion estimation entails building theories and algorithms capable of processing how the camera and the subjects move in flip-book-style data such as video, determining where they will be next, and then implementing these through programming.

Applications in various fields and illusory motion

Research into motion estimation is expected to have applications in autonomous driving and other diverse fields. In fact, the results of my own research have been used in video data compression technologies. In video data compression, the different flip-book images of a video are not compressed frame by frame; rather, by identifying differences between one frame and the next, the total data size can be reduced.

I have shown that we can use motion estimation-based predictions to minimize differences between frames—so further compressing data volumes and improving compression efficiency.

In my research, I am also exploring the use of motion estimation to predict the movements of clouds. From sequences of satellite images and ground-based video, it is now possible to estimate and predict the motion of clouds in three dimensions. Such research may well be of use in forecasting solar power generation.

Ever since I was a university student, I have been fascinated by illusory motion. One example is the striped poles found outside barber’s shops in Japan and elsewhere, in which the stripes appear to travel up the pole. At present, I am involved in research aimed at explaining these illusions using motion estimation theory.

I am also looking into new camera data processing technologies capable of grasping only aspects of an image that have changed at ultra-high speeds. These cameras are known as “event cameras,” and are expected to feature on drones and other airborne vehicles of the future.

Invisible movement in martial arts

My activities center on basic research aimed at improving theories related to video processing, and I believe that motion estimation is a field in which I can work together with specialists from fields as wide-ranging as engineering and the humanities. Such collaborations often result in unexpected developments, which can be both stimulating and fruitful.

AI-driven approaches—using machine learning to generate motion data—are increasing in the field of motion estimation, too. But since it is difficult for AI to understand the theories that underpin different phenomena, I believe that basic research remains essential.

As a researcher, it is vital that I develop an interest in things I don’t understand, and that I continue to think about why things are the way they are. Even in my everyday life, I am constantly thinking about motion. One of the things that intrigues me is how in the martial arts, some movements are almost invisible. As a teacher, I hope that my students come to understand the pleasures that thinking can bring, too.

The book I recommend

“Flatland: A Romance of Many Dimensions”
by Edwin Abbott Abbott, Japanese translation by Kaoru Takeuchi, Kodansha

Flatland is a science fiction novel in which a square that inhabits a two-dimensional world is visited by an inhabitant of a three-dimensional world, who teaches the square about worlds of various dimensions. The novel is a satire on the class and gender-based hierarchies of the Victorian era. It transcends the domain of science and technology, and is a text befitting of Sophia University. I hope it inspires readers to consider the question: what is four-dimensional space?

Interviewed: June 2025