Depth Estimation with Tilted Optics Using Split Color Filter Aperture

Camera-based depth estimation methods are useful for various automobile applications. For example, depth estimation methods based on stereo cameras triangulate the depth between objects and cameras. However, they are susceptible to baseline distortion owing to vibrations, aging, and thermal expansion. On the contrary, depth estimation methods based on a monocular camera, such as depth-from-defocus, use a focusing mechanism or spectroscopic optics to calculate the depth from the difference of defocus images. However, because the change in sharpness, which is the inverse of defocus, is slightly in the depth direction, it is unsuitable for deep-range estimation. Therefore, we are developing depth-estimation methods that use a monocular camera with special focus images obtained through tilted optics, which have no baseline distortion problem and have a deep estimation range. When tilted optics are used, the plane of sharp focus (POF) appears tilted, while the depth of field (DOF) lies and increases toward the depth direction. Moreover, to allow for the use of only one image-sensor, a one-shot, we studied a depth estimation method using tilted optics with two different aperture sizes for each color wavelength by introducing concentric color filter apertures. However, as with conventional methods, the Gaussian sharpness model does not strictly fit to the actual sharpness, resulting in reduced estimation accuracy. In this study, we propose a method for high-speed and high-accuracy depth estimation that involves adding a split color filter aperture to a tilted optic and using a different sharpness model for each DOF side.