DEPTH ESTIMATION FROM FOCUS AND DISPARITY

This paper explores how focusing and defocusing can be used in combination with stereoscopy to enhance the accuracy and speed of depth estimation. The proposed method for combining focus information with stereo is inspired from how the human brain perceives depth using both. The outline of the method is to first estimate depth from focus using a high focal length camera. Next, we perform disparity estimation from the stereo images by only searching in the neighborhood of the disparity from focus (derived from depth from focus) for matching features. The matching is faster as the search range is only a small neighborhood of the expected disparity map. It also increases the accuracy because searching over a small range of high confidence rules out the possibility of wrong matching, especially if there are multiple areas with similar features. Depth from focus can be coarsely approximated by depth from defocus using multiple cameras focused at complementary distances. Hence, the proposed method also works with depth from defocus.