A Method of Extracting High Rersolution 3D Information Using the Stereo Camera and Depth

This paper addresses a method of extracting depth information using a stereo camera and depth. We use Kinect developed by Microsoft and Prime Sense to generate range images. Range images are often used in 3D modeling. Because range images are convenience for 3D shape estimation and view point move. Various methods are known generating a range image. However conventional systems still exhibit limitations in cost and fundamental problems. By the launch of Kinect as consumer device, we can generate range images easily. But the range image generated by Kinect is low resolution, and has some noise. In this study we use a range image generated by Kinect as reference point for a stereo matching. Using these points we can eliminate the problem of occlusion. And using the noise of depth we limit the aria of finding corresponding points. First we determine the accuracy and the noise of depth generated by Kinect. Second we determine intrinsic parameters and spatial relations of Kinect and stereo cameras. Third we transform the depth to overlap each view of stereo cameras and estimate the occlusion area. Finally we find corresponding points in limited area using the transformed depth information