Three-dimensional imaging technology has been widely used in various applications such as on-line inspection and gesture recognition because of its ability to detect 3D information of the scene. Current, depth camera and binocular stereo vision are two main techniques to gain depth image. In this thesis, we present a combination of the two main techniques with the purpose of detecting high-precision depth map of close scene fast.The overseas and domestic research status of depth camera and stereo vision is described, and so are the principles and characteristics of the two techniques. A dual-mode camera which can be used to take stereo images or depth maps is developed in our work. Under the constraint of the depth map with the same spatial resolution as the stereo pairs, expected disparity of each pixel is limited within a narrow search range. The experimental results demonstrate that it can speed up the process of stereo matching and reduce the error in matching effectively. The coarse depth map is also used to fill the invalid region of the matching result.In this article, the process of system calibration is described in detail. A new matching cost based on the joint probability distribution of intensity between the two stereo pairs is proposed. Simulation results show that this matching cost can produce results similar to non-parametric matching cost, and with the value of further research.Finally, the summary of this thesis and discussions about the further research are proposed. |