Abstract
There have been a number of researches done on methods for computing three-dimensional information from images taken with different focal settings. However, due to the bottlenecks in the speed of both optical system and algorithm speed, these methods are rarely employed in a case where the scene is dynamic. This paper describes new computationally inexpensive method for estimating three-dimensional motion of feature points. The method computes optical flow between two sets of images where the focal plane of each image in a set is different. Its validity in real world case is demonstrated on an image sequence that is captured at 240 fps using high speed liquid lens, called Dynamorph Lens, allowing each image to be captured at different focus.
