3-D shape reconstruction from an image sequence


Original Image.

Optical Flow.
The original image sequence was taken from the roof of a car going straight ahead. The optical flow was calculated using two adjacent frames.


Reliable Optical Flow 1.

Reliable Optical Flow 2.
Optical flow was selectec by its relability. The covariance matrix of each optical flow vector was calculated. The vectors were selected by the largest eigenvalue of the matrix. The above two results are different in the threshold value.


Reconstructed 3-D Shape.

Reliable Region of The Shape.
The 3-D shape of the environment was reconstructed using the optical flow and its covariance matrices. The motion parameter of the camera movement was also obtained, which is a byproduct for the aim of shape reconstruction. Since our algorithm yields reliability information (covariance matrices) of the all results, we can select reliable area of the shape as shown in the lower image.