Estimating the 3D direction of a translating camera from a single motion-blurred image

We propose an algorithm for estimating the 3D motion direction of a camera that undergoes a pure translation. This algorithm exploits a single blurred image, and recovers the 3D translation direction thanks to an accurate analysis of the motion blur, which is characterized by rectilinear smears whose directions and extents typically vary throughout the image. The core of our algorithm is the estimation of the direction of these smears within small image regions that are automatically selected according to the image content. The algorithm has been successfully tested on camera images and extensively validated with different amount of noise in the images.