We study some topics for reconstructing 3-D shape of 360 degrees
view from video image.

1. Incorrect feature tracking detection

We detect incorrect feature point trackings by fitting an optimal affine space to input feature trajectories. We do random sampling of the feature trajectories and fit an optimal affine space. By computing the residuals of the fitted affine space, we detect incorrect feature point trackings. We also estimate the correct positions for the detected trackings so as to minimize the distance from the fitted affine space.

2. Triangular mesh integration

We propose a new method for integrating triangular meshes
independently generated from different images. We detect
overlapping triangular patches and correct inconsistencies among
them by re-generating triangular patches.

We study high accuracy ellipse fitting and ellipse detection
methods. In order to detect an ellipse from an image, we extract edge points from the image and fit an ellipse to them. In this case, the accuracy of ellipse fitting deteriorate if the input edge points include outliers. For this problem, we propose a new method for fitting an ellipse to the edge points even if they include outliers.

We also develop a mixed reality sysytem using a circular marker.
In this system, we detect a circular marker using our proposed
ellipse detection method and impose virtual objects on the image
by computing the camera pose and position from the detected
marker.