3D Scene Acquiring And Modeling Based On Images

Non-contact measurement based on image has many merits that traditional measurement methods don't have. So it has been attached importance to and has been applied in many cases. More and more researchers have entered the field and been devoting to it. Camera calibration is a long lasting research topic in this field and photogrammetry doesn't need special equipments any longer with the popularization of high resolution and non-measurement CCD digital camera. How to achieve high-precision calibration of those cameras is one of the research points of this paper. The camera model is adopted from [Weng 1992] and high-order coefficients of various distortions have been considered to improve measurement precision. Eventually camera-parameters can be found by solving minimization with Levenberg-Marquardt Algorithm and tests show this process is fast and high-precision calibration has been achieved. So we can change the widely known viewpoints that only radical distortion can be considered in measurement with high-distortion camera.The notions of Dynamic Subsection Calibration and ROP factor will be given in this paper. That is localizable camera-parameters can be got by considering only a local region around the target points and the camera-parameters based on the whole image can be used as a good estimate. Tests show that this method is very effective and very good results can be achieved especially only considering low-order coefficients of distortion.3D model construction from multiple views can be accomplished using the matching points of multiple views. Translation vector and rotation matrix can be got by non-linear least squares algorithm whose initial parameters are gained by linear least squares. Our test shows that this method is simple and effective.A model of scene has been reconstructed successfully by using above technology.