Dense depth map generation using a single camera with hybrid auto-focusing

Abstract

This paper proposes a novel depth map generation method using a pair of stereo images obtained from a single sensor. The proposed method generates a dense depth map as follows: i) sampling of the stereo image pair using photodiodes equipped with the black mask, ii) feature extraction in low light and noisy environment, iii) matching of the stereo images using motion estimation, and iv) matting of the matched image. Because stereo images are obtained using only one sensor, this method can estimate the depth information without special devices.Keywords-depth map, hybrid auto-focusing, single sensor, stereo matching. I. IntroductionDense depth estimation has been steady studied to obtain three dimensional (3D) information and many products can generate a depth map. For example, infrared camera systems are used to change program channels in the smart-TV with gesture recognition. Distance between vehicles is also estimated and measured by stereo camera.The one type of depth estimation method uses special devices such as infrared sensors and stereo cameras. An infrared sensor with time-of-flight system can quickly measure a distance between a camera and an object. However, this device has a limited range of the distance. The other methods generate the depth map using a stereo image set from a stereo camera or multiple color-filter aperture camera [1]. Despite an accurate depth map with fast computational time, the method needs the high cost to generate depth map.There is another type to obtain the 3D information. This type estimates a vanishing point or point spread functions (PSFs) from a single input image. The vanishing point is generated by projecting a 3D space on the 2D image space. In additional, the PSFs in an image have different values because the distance between the camera and the object is different with the focus state. In this type, the depth information is computed by only one sensor. However, these methods have less accuracy than the former type because it is difficult to estimate the accurate parameters for computing the vanishing point and PSF.To solve these problems, this paper proposes a dense depth map generation algorithm using sensor-based phase