Three Dimensional Reconstruction Algorithm Based On Stereo Vision And Depth Sensor

As the rapid development of computer vision technology, it has been widely concerned in many fields. The three-dimensional reconstruction technology is a hot issue in the visual research, applying in computer animation, medical image processing, virtual reality and aerospace and other fields. Depth acquisition is a prerequisite for three-dimensional reconstruction, but in some applications of high accuracy requirements, it needs more accurate depth information.At present, there are two ways to obtain the depth information: passive acquisition mode and active acquisition mode. Passive acquisition is mainly get multi pictures of the scene through the multi cameras to get the depth information of the object in the scene, through stereo matching. Active way through active lighting of the depth camera on the scene, projection of special light to get depth information.Passive stereo matching method to obtain the depth is affected by the scene, when the image with the weak texture region and the area of the block, it can not get accurate depth information. And the depth map of the active depth sensor is not affected by the factors of the scene, in these special areas can have a good performance, but the depth sensor has its own shortcomings, such as high noise, low resolution, and in the transparent area and special material object is also very limited. Therefore, this paper decided to use the combination of passive and active mode to access depth information, to get the fusion depth map of stereo vision and the depth sensor, and use the depth map to make three-dimensional reconstruction. The main contents focused by this thesis are as follows:In order to better fuse the depth information of stereo matching and depth sensor, the multi cameras system calibration is needed. We first analysis the model of camera imaging principle of camera calibration, and then use the classical calibration method based on planar camera calibration system.The principle of stereo matching is studied, and the general process of stereo matching is summarized, and the confidence propagation algorithm based on Markov is studied.By integrating the texture and depth information into the Markov-theory-based maximum a posteriori probability framework(MRF-MAP), to model and optimization calculation the high precision depth map is modeled as an optimization problem, and the maximum a posteriori(MAP) is obtained by means of the belief propagation(BP).With high precision based depth map fusion, using plane projection method for 3D reconstruction, using the OpenGL graphics library for rendering, to get realistic 3D scene reconstruction scene finally.