단안 카메라 기반의 휴대단말기기에서 무안경 3D를 위한 사용자 위치추적 시스템

Abstract

Recently, the interest for 3D has been increased. Therefore, many devices for viewing 3d contents that require special glasses are invented. But, wearing special glasses is uncomfortable to enjoy 3d effect because of the feeling of irritation. So, techniques that do not require such devices have been developed to meet user needs. However, due to the characteristics of the display, such technique allows user to view 3d effect only at certain locations. To deal with that problem, estimating real world location is necessary. If the user location in real world is estimated, the barrier of the LCD display can be controlled to provide user the best 3d effect. There are two kinds of method to estimate real world position. The first one is using a stereo camera. It is easy to estimate real world position using stereo camera, but computational cost is too high for a portable device, a limited hardware system such as a cellular phone. The method using a single camera may be fast enough to be applied for the limited hardware system but with some constraints. The cost as well as the computational complexity for the system may be low, but so does the accuracy of the system. The depth information cannot be obtained accurately using the single camera alone. So we propose an algorithm that runs on a limited hardware to meet not only the speed constraint but also simple and inexpensive solution by using a single camera. To increase the accuracy in the real world coordinate system, we use two key ideas: one is to estimate face rotation angle by calculating 3d rotation angle in a single image. Then, adjust the distance between left and right canthus depending on the rotation angle of the face. The other is by using a mapping function to calculate the real world coordinate. To make a mapping function, a simple calibration method is used to find the relationship using an ordinary object that can be found everywhere such as a name card. Finally, a mapping function is generated to measure the distance. The experiment results show that real world coordinate can be found with good accuracy using a single camera. Also, our algorithm runs at 15ms which is enough to apply for mobile devices. Also, the accuracy is good enough to display left and right-images for auto-stereoscopic 3D systems.