Research Area

Robot navigation as motion geometry

We consider robots as geometric objects moving in three-dimensional space. Robots obtain geometric information of their envirnments. We develop algolisms for robot controlling based on time change of geometry features, such as points, lines, planes and surfaces captured by cameras mobiled on the robots.

Humanoid robot (left),  Mobile robot(right).

Image Analysis on Surfaces

Images captured by catadioptric and dioptric camera systems are expressed as functions on surfaces. We deal with computer vision, pattern recognition and signal processing of images on surfaces and manifolds. Multiple view geometry with omnidirectional cameras is also our main interests.

3D Reconstruction using Range Sensor

Range sensor mesures three-dimensional information. Mobile range scencer is our current interest. We deal with geometry reconstruction by mobile range scencer. Surface reconstruction and mosaicing using geometric constraints are the practical issues.

Small range sensor with pinhole camera (left), range sensor for wideview (right).

Euclid reconstruction from discrete data

Reconstruction of Euclidean shape from discretized data has long history. Designing mathematically and geometrically robust algorithm is our main aim.

Scale Space -- Image analysis at all levels of resolution

Image detail, quality and information depend on resolution or scale of the image. The aim of our study is to clarify qualitative properties of image throughout the scale.

We develop the scale-space theory and analysis techniques, which provide us with the image properties, such as hierarchical relationships and configurations of image segments. Potential applications are image classification, identification, temporal segmentation of motion image, etc.