Coregistration of terrestrial lidar points by adaptive scale-invariant feature transformation with constrained geometry

To obtain an accurate 3D model of the real world using a laser scanner, point clouds should be registered precisely. To increase the registration accuracy, the authors used wavelet based noise removal filters on the point cloud data, and extracted feature points from intensity images using the SIFT (scale-invariant feature transformation) method for two overlapping point clouds. These feature points were then used for a corresponding point matching to obtain a rigid body transformation matrix by an iterative technique. nitial CTNC (closest-to-next-closest) ratio of 0.4, points were extracted and the transformation matrix was calculated. Under this geometric condition, the CTNC ratio was increased to obtain more points for matching. Then, the transformation matrix was recalculated with these points, giving more reliable results. The outliers were removed by random sample consensus (RANSAC) processing. sure and analyze the performance of our approach in pairwise registration, additional transformation parameters were computed using the Polyworks commercial software. Comparison of the two methods showed no significant difference in mm level. In the final stage, all the scan data are rapidly adjusted using global registration, due to a small number of accurate control points. Thus, the proposed coregistration method can be used to obtain fast 3D modeling results on construction sites where registration targets cannot be installed.