Research Of3D Reconstruction Based On Image Sequence

Three dimensional (3D) reconstruction has become an important research topic in the field of computer vision. Recently, image-based3D reconstruction techniques gradually become a research focus in this field. The scene or target objects from general cameras and corresponding cameras’ parameters are used to estimate the3D information of space points. This method is of lower time complexity and can realize semi-automatic or even automatic modeling. Because of its advantages, this technique has been widely used in mobile robot navigation systems, aviation, medical imaging, industrial automation systems and many other fields.In this paper, the multiple view images of a target object are used to achieve the3D reconstruction result. The reconstruction procedures are mainly studied in the image preprocessing, feature extraction and matching, camera calibration and3D reconstruction. This paper is mainly divided into the following sections.Image preprocessing. To obtain more desirable reconstruction model, image preprocessing methods are usually adopted for image sequence. The algorithms of removing salt and pepper noise from images are studied thoroughly in this paper. An adaptive filter based on GA-BP neural network is proposed in this section. Experiment results show that the proposed algorithm significantly outperforms the others and efficiently removes salt and pepper noise from digital images without distorting image details and textures.Camera calibration. The principles of camera calibration are introduced in this section, and then a calibration method based on dot template is used to achieve the camera parameters. This method automatically detects the points of the pattern features, and the experimental results show the robustness and the accuracy.3D reconstruction. In this section, the multi-view stereo reconstruction method based on the algorithm of structure from motion is introduced. This method is widely applied and doesn’t need calibration objects. The accuracy of this method is verified by comparing with the results of dot template calibration. At last, practical image sequences are used to reconstruct3D models of different targets.