Multiple Views

In this group, we concentrate on 3D reconstruction by using multiple cameras and sensors. In order to generate high quality 3D model, we utilize computer vision technique such as stereo vision, camera calibration and image segmentation.

Free viewpoint video synthesis using multi-view depth and color cameras

This is a system for generating free viewpoint videos based on multiple depth and color cameras. The main contribution is the capability to enhance the depth maps to increase the quality of the result by performing an upsampling with reduced amount of noise.

Depth map cleaning

Development measuring methods for table tennis balls in a match

Nowadays, data becomes to be used for match analysis in practical scenario in sports. However, there are some problems, such as enormous amount of time, and easy way to measure the velocity or rotation of balls.

In this research, we are developing (1) a method for reconstructing the 3D trajectory of a ball, (2) a method for measuring the rotation of a ball, and pursue to establish basic methods that makes easy to use data in practical scenario in sports.

Tracing a ping-pong ball path

Analysis using video in training of rowing

In rowing, the index that is important is a time mainly. However, in this research, by paying attention to the movement of oars in the video in training, and easier to see the movement of oars by the moving image processing, to consider how the motion oars whether associated with the boat speed. By utilizing the video in training, it is assumed that it is possible to analyze the motion of a natural environment usual.