Research On Registration Of Multi-view Vision Measurement Data

As one of the key technology in vision measurement, the precision of the multi-view registration will directly affect the overall accuracy of the measurement. Currently widely used based on registration point mark method is easy to implement and other auxiliary equipment splicing method uses the low equipment cost advantage relative. However, there are also mark point precision measurement and error accumulation problems in continuity. The mobile template and using the bundle adjustment algorithm is used to measure a single regional and global overall template sign point coordinates and the measurement position parameters were optimized, and with the optimized parameters of measured point cloud of registration, the point cloud registration fusion fairing.Is designed in this paper made a mark point template, presents object surface point cloud measurements were made with a mobile template combination of point measurements of registration measurement scheme of indications, and put forward the measurement and data processing of the whole process, which includes calculation of three-dimensional coordinates of the point mark, with different angles of the same name sign point matching, camera external parameters calculation, bundle adjustment optimization and point cloud registration and so on.In based on bundle adjustment of multi view registration optimization, this paper adopts binocular stereo vision method calculation template landmark coordinates and through different angles between the template sign point coordinates of the matching calculation of the perspective camera external parameters, and this camera external parameters and die plate signs point coordinates for bundle adjustment of the initial values for iterative optimization. Next, through the different measurement positions of the template coordinates of the template after the optimization of the optimization of the coordinates of the coordinates of each measuring point under the coordinates of the coordinates of the template into the same coordinate system. Then the global beam method adjustment is carried out to optimize the transformation relationship between the different measuring positions. The optimization of the camera parameters of the point cloud registration, and the results were compared with the three coordinate measurement. The results show that registration measurement of average error is 0.025 mm, standard deviation is 0.022 mm to verify the feasibility and accuracy of the registration method.In point cloud fusion, we first use the index algorithm to pre-process the point cloud, and filter out the outliers. Then, the fusion point is calculated based on the moving least square curve fitting method. The results show that this method can eliminate the phenomenon of point cloud stratification caused by registration error, and make the whole point cloud more smooth.