立体视觉是计算机视觉领域的经典内容，一直是广大学者大力关注和研究的内容。立体视觉的二个关键问题是立体视觉标定和立体视觉匹配。立体视觉标定追求标定的高精度和简单性；立体视觉匹配则要在弱纹理、遮挡、噪声等因素的影响下，求解对应点问题，是计算机视觉领域最具挑战性的问题。本项目针对两个关键问题，研究嵌入式标定立体视觉方法，目的是实现在线全自动立体视觉标定和场景高精度彩色-深度图像（RGB-D）恢复。嵌入式标定是指采用光学镜组把位于场景外的标定装置图像嵌入到场景图像中，等同于在场景内放置标定装置。本项目的研究内容包括嵌入式标定立体视觉系统设计、嵌入式标定立体视觉系统的自动高精度标定方法、嵌入式标定立体匹配方法和深度估计方法、深度图增强方法等。本项目预期研究成果可以促进计算机视觉方法的发展，并推动计算机视觉技术的广泛应用。

Stereovision system is one of the most attractive topics in computer vision. It can be widely used for many applications, such as 3D scene understanding, obstacle avoidance of autonomous robotic vehicles, and et al. The two key problems of stereo vision are camera calibration and stereo matching. This proposal proposes to use embedded calibration device to realize online automatic strong calibration to obtain high precision and high resolution depth map. Under this framework, we plan to investigate the design of embedded calibration stereo system, the automatic calibration method for the embedded calibration stereo system, the stereo matching method for the embedded calibration stereo system, and the depth enhancement method for the depth map obtained by the embedded calibration stereo system. This proposal poses significant importance on proposing new theroies and methodologies on stereo vision calibration and stereo matching, and also beneficial for developing more technologies and methods for wide applications of stereo vision systems.