Research On 3D Marker Reconstruction Methods From Multiple Uncalibrated Images

With the fast development of computer technology, network technology and information technology concerned in recent years, people are in growing interests in obtaining 3D information of the real world via computers, thereby setting a higher threshold to the applications of 3D reconstruction technology. However, most of these methods are based on camera calibration technologies directly or indirectly, which block the way of the 3D reconstruction to practicability and universality to a certain extent. To solve the actual application problem, this dissertation researches on how to avoid the camera calibration process and the technology of automatic reconstruction of scene from the view of Structure from Motion and researches on several key technologies in the 3D marker reconstruction process based on uncalibrated images. The main works are listed as follows:(1) An improved factorization method for structure from motion is proposed under orthographic projection model. This method achieves the goal of fast efficient reconstruction of object structure through simplifying the camera projection model. It avoids the invalid situations of decomposing non-positive-defined rectification matrix in conventional method and obtains results with higher stability and accuracy compared with the conventional method.(2) An iterative factorization method for structure recovery is proposed under real perspective projection model. The method simplifies the camera intrinsic parameters model and achieves the high efficient reconstruction by iteratively rectifying the camera motion matrix. The reconstruction is satisfying and we can obtain the camera focal lengths accurately as well.(3) A 3D marker reconstruction scheme based on multiple uncalibrated images is designed. At first, several predesigned markers and standard ruler are placed at key points of the scene, and then the camera shoots at the scene from multiple angles and at last, the inter-structure of the key points is reconstructed by structure recovery algorithms. Taking the structure of markers adopted into consideration, detection and location method is designed respectively. Meanwhile, a marker matching scheme of using color matching to guide the epipolar matching is designed. It first uses the color matching of 8 special markers to build the epipolar geometry correlation of two images, then guide the rest ordinary markers to match their location based on epipolar geometry and lastly achieves matching all the markers exactly. Experiments show that this scheme can match all the markers at a very stable level.