In recent years, omnidirectional cameras, which can capture 360-degree panoramic photographs and videos in all directions, have been used in a variety of industries. A representative example is the RICOH THETA. We are conducting research that takes advantage of the wide-area imaging capability of omnidirectional cameras, including high-precision three-dimensional measurement and self-localization for autonomous robots.
RICOH THETA
360-Degree Photograph
from the 5th Floor of Building No. 3
With the rapid development of robotics, robots have become increasingly familiar and closely connected to everyday life. Recognizing the surrounding environment is an essential task for robots to perform work. Omnidirectional cameras are inexpensive, can capture images over 360 degrees, provide a large amount of information, and can acquire measurements with a single shot. For these reasons, we designed a stereo measurement system in which conventional cameras are replaced with omnidirectional cameras. However, binocular omnidirectional stereo has problems such as reduced accuracy along the epipolar-line direction and blind spots. To address these issues, we perform stereo measurement using a new configuration that adds a third omnidirectional camera.
-関連論文-
Equirectangular Image
of a Classroom
3D-Reconstructed Classroom
High-precision three-dimensional measurement is important for maintenance of large-scale facilities and infrastructure. We propose an instrument and measurement method that combines an omnidirectional camera and a ring laser to achieve dense measurement while also keeping costs low.
Indoor Scene
3D Shape of an Indoor Space Measured Using Laser Light
Full-circumference laser Projection
GPS may not function correctly in places where radio waves do not reach or where they are reflected, making it difficult for a robot to estimate its position accurately. Therefore, we use an omnidirectional camera, which can capture a wide area, to detect doors in the surrounding space. The camera position is then estimated by comparing the detected information with a pre-prepared two-dimensional map that includes object height information.
Door Detection Using Deep Learning
Prior Map