Research On 3D Reconstruction Based On Synthetic Vision

3D reconstruction is an important research topic in computer vision. It is mainly used in autonomous robot navigation, target identification and real-scene remodeling. For needs of UAV autonomy landing, this paper proposed a framework of 3D terrain reconstruction processing, focusing on theoretical analysis of synthesized vision and experimental data processing.On the based of stereo vision, this paper analyzed the error accuracy of 3D structure reconstruction. And a space-time processing algorithm for monocular synthesized vision was proposed with matching-based method (SFM, structure from motion) and non-matching method(plane parallax method) respectively. SIFT (Scale Invariant Feature Transform) was extracted for this algorithm's robustness. Through theoretical analysis and experimental data processing, 3D reconstruction was achieved from binocular, monocular short baseline to long synthesized baseline.The innovation in this paper is that the plane parallax is used for the single-camera three-dimensional reconstruction. Camera internal parameters were obtained by calibration. Under the brightness invariance constraint, the rotation and translation parameters of a camera and scene structure parameters are both estimated by Newton iterative optimization algorithm. Then 3D structural is realized. The algorithms improved the accuracy of 3D reconstruction, and eliminates the middle processing of the image matching.