| With the development of technology, Reverse engineering is used more andmore widely in modern manufacturing industry. Point cloud data registration isnot only one of the key technology of reverse engineering, and it is also moreand more widely used in the CAD model reconstruction based on point clouddata, error detection, target recognition and other fields. However, the presentautomatic registration algorithm is limited in the practical application because ofits poor stability and low registration efficiency. Taking automobile cover partsof the point data as the research object, the complete point cloud data wasobtained through different registration method.The content of the thesis mainly includes:1. The point cloud spatial topological relations were established rapidly bythe space grid strategy. Gaussian mapping was established by using the leastsquares fitting method to extract the point cloud surface method of vector. Thevector information of the point was extracted by using the least squares fittingmethod of curved surface. Normal curvature mapping were established with thevector information of the point.2. According to the characteristics of feature surface in the Gauss mappingand normal curvature mapping, this paper puts forward the feature surfaceextraction methods based on the Gauss mapping and normal curvature mapping.3. The expanded Gaussian sphere was established by the normal vectorinformation and curvature information. The matching template with rotationalinvariance was established. The template was used to searching correspondingpoint. In this way, the initial registration for point clouds based on the expandedGaussian sphere can be realized. Finally, the ICP algorithm is adopted toimplement precise registration for point cloud data. Aiming to the shortcomingsof the Gaussian ball registration method based on extension, this paper proposedbased on the registration method of local coordinate system of point cloudregistration method. |
PDF Full Text Request