Binocular Stereo Image Acquisition Based On ARM And Research On Reconstruction Algorithm

Changing the flat panel display to three-dimensional visualization can not only give people a real senses, but also will make industrial designs, measuring and modeling for three-dimensional world more scientific and accurate. Therefore it is valuable and significant to study the three-dimensional vision technology. In this paper, four major aspects of binocular stereo vision system have been studied. These are binocular stereo image acquisition, camera calibration, stereo image matching and image reconstruction. Aiming at developing three-dimensional vision applications in multimedia handheld devices, this topic has selected the image acquisition program based on dual cameras and ARM platform. Codes have been programed to drive the dual cameras and capture images. What’s more, a method which can capture stereo images by single lens has been proposed to relief workload of camera calibration, streamline the camera device, and reduce costs. In the aspect of camera calibration, transformation relations between the four coordinate system, and three major camera calibration methods, such as traditional calibration, self-calibration and active visual calibration method, have been analyzed. Then the Zhang-zhengyou calibration method has been used to obtain the internal parameters of the two cameras used in this program. In the aspect of stereo image matching, the main problems of this field and three important methods, such as region matching, feature matching and phase matching, have been analysed. Research work focuses on the census transformation matching method, which is one kind of region matching methods. It has been improved through combining the feature detection.In this way,not only does it keep the dense disparity of the region matching method, but also wins the accuracy and stability belonging to the feature matching method. Moreover, a real-time algorithm has been introduced to the region matching method. In the three-dimensional reconstruction work, a model of parallel linear binocular camera has been used. With the well matched points by the improved census transformation, three-dimensional image reconstruction has been carried out on the personal computer with MATLAB. The main contours of the reconstruction image is consistent with the real objects, and has a certain depth.The experiments in this paper have achieved the desired results. They have verified the feasibility of this paper’s methods used in binocular stereo image acquisition, camera calibration, stereo image matching and image reconstruction. The research results are of some practical value.