Current Research

Brain Analysis, Realization and Support

OUTLINE
 The human brain is an excellent natural machine that performs important information processing such as seeing (vision), listening (audition), understanding languages, and motion controlling.
 We focus our interests on the modeling of these information processing (human information processing). From an engineering perspective, we analyze the mechanisms of human information processing and realize them on computer systems such as robot vision, robot audition, and natural language processing. Furthermore, understanding these functions also helps enhance the field of rehabilitation engineering, which aims to realize sensory substitution systems for persons with visual or hearing impairments.
TOPICS
(1) Analysis of Human Information Processing
 The analysis and modeling of human information processing is important not only for the pursuit of scientific truth but also for artificial realizations of biological systems. We model and simulate human audio-visual information processing based on physiological and psychological knowledge. So far, we have conducted the researches on analysis of the binocular gazing point, perceptual models of figure and transparency, models of generating subjective contour, and emotion embedded in speech, Now we are studying audio-visual event corresponding, which is the basis of sensor fusion and the acquisition of object knowledge.
(2) Robot Vision and Audition
 Humans understand their environment by using their eyes and ears. They can localize a sound source location (source localization) and recognize the sound (source discrimination). Furthermore, they can listen to one sound among other simultaneously existing sounds (source separation). Humans also understand the back-and-forth relation of multiple objects (depth perception), an object's 3D shape (shape recovery), and recognize objects (object discrimination).
 So far, we have conducted the following researches : a sound source localization system inspired by mechanisms in living organisms ; basic methods for blind source separation ; extraction of inner contour based on the subject contour generation model ; a method for detecting object overlapping based on the transparency perception model ; 3D shape recovery from shading information ; and the extraction of character strings in a scene image. Currently we are studying source discrimination in indoor/outdoor environments, blind source separation in the real world, the extraction of character information in an outdoor scene image, and the estimation of emotion from both speech and faces.
(3) Rehabilitation Engineering (sensory substitution)
 The human's fundamental functions are seeing, listening, speaking, and motion controlling. Using these functions, they communicate with others, systems and environments. Impaired persons, however, have difficulty understanding and communicating with their environment. We have developed a diagram recognition system for blind persons, a system for acquiring spatial awareness, and a system for imparting character information to the blind. Now we are engaged in detecting such emergency sounds as horns.
 The researches described above are applicable to audio-visual scene monitoring and the realization of artificial eyes and ears necessary in autonomous robots. Our research principle is to extend human ability and aid human activity.