Development Of Multi-camrea Calibration System

3d face is widely used in virtual reality, entertainment, medicine, pattern recognition, etc. Usually 3d face reconstruction method can be divided into based on structured light method, the method based on 3d laser scanner, facial 3 d reconstruction method based on the medical CT images, 3d face reconstruction method based on the deformable model and 3d face reconstruction method based on the camera images. The method based on the camera images is recently the most popular research direction. It usees the principle of stereo vision, which simulates human eyes treatment of scenery. Binocular stereo imaging by using two relatively fixed cameras, from different angles and at the same time to obtain the same two images of the object to be reconstructed, then by calculating the corresponding points between two images, and through reprojection matrix to obtain a 3d point coordinates. But because of the limitation of size and depth of field lens and field of view, two cameras can only get a part surface of the 3d object, and because of the shadow of the environmental illumination, the object with self-occlusion, generally a binocular stereo measuring system can only get a part surface point of view. So, for reconstructing the overall surface of the object, it is frequent to capture images from a different point of view using multi-camera, and then will be to merge measurement results to get a complete surface reconstruction.The aim of this thesis is to establish a set of suitable three-dimensional reconstruction imaging system for the human face. Because different viewpoint reconstruction result is not the same coordinate system, need to study how to merge these measuring data under different angle of view. Although it can be done by the registration directly to calculate the coordinate transformation between measurement data, but in stereo vision, this can be solved through the multi-camera calibration method.Camera calibration of stereo vision is the most basic, the most important part, although a lot of monocular or binocular camera calibration methods are put forward, but for the multi-camera calibration, is a few. Generally they can be divided into two categories, based on the calibration object and camera self-calibration. The former usually choose one dimensional, is higher accurate, but have restrictions on usage scenarios; And although self-calibration method is suitable for various scenarios, but the general lack of robustness, precision is lower. In the third chapter puts forward a kind of 4-camera calibration algorithm, the method based on the plane calibration algorithm, easy to use, and the algorithm can be extended to more camera, any two cameras can form a stereo pair.The significant result about this thesis includes two aspects:First, a face 3d imaging device has be builded. The device is based on passive vision stereo imaging system, and is composed of four high resolution camera Canon EOS600 D. Imaging process can be outlined as: the cameras around are divided into two groups, respectively, based on the principle of stereo reconstruction of left and right sides to reconstruct the 3d face; Then left and right sides are merged to obtain the final 3d face.The second is finished about 3d face merging. The key idea is that firstly the left and right stereo pairs are independently calibrated, and are transformed to the respective calibration correction of three-dimensional space. Then two correction space are considered as a stereo pair, thus the spatial relationship can becalculated by another stereo calibration process.In the end, the synchronous(soft synchronization) capture software based on the Canon development kit ED SDK 2.11 was developed four cameras; Using Zhang’s plane calibration algorithm, based on OpenCV- 2.4.3, the software of 4-cameras calibration is implemented. The real data test shows that the desired goals.