Surface Deformation Monitoring Using 3D Laser Scanning Technique

As the recently developed high-tech, three-dimensional laser scanning technology can access the three-dimensional data of the observation object's surface rapidly, accurately, intensively and uncontactly. It has changed the traditional single-point deformation monitoring technology, transformed "point measurement" method into "shape measurement" method. Therefore, the three-dimensional laser scanning technology applied into deformation monitoring has important theoretical and practical significance. Compared with traditional measurement techniques, three-dimensional laser scanning technology increases the degree of extracting information automatically. It can collect data of survey area intensively and comprehensively without layout point, equivalent layout a high density reticular deformation monitoring network, which provides the more comprehensive measured data for deformation monitoring research.Point clouds model obtained by three-dimensional laser scanning technology of large terrain area is based on discrete sampling points, and cannot be directly applied in the field of deformation monitoring. We need to reconstruct terrain surface of the clouds data using mathematical or figure description, and then extract deformation during different period. This paper discussed surface reconstruction of the massive point clouds data using neural network. This method reconstructed surfaces of the point clouds data obtained in two periods and sinking surface of the two period point clouds data. It can not only achieve the entire region subsidence directly, but can access the sinking value on any point.The findings of this paper are summarized as follows:(1) By using parameter optimization method, we unified coordinate system of the two period point clouds data in Matlab accurately which provides base for the deformation study during different period.(2) We proposed to reconstruct surface using least squares approximation of B-Splines, and implemented in Matlab. The results show that fitting accuracy of this method is very high, but it is time consuming because each point must be computed least squares. If use less points, the fitted surface's continuity is poor, and lost a lot of detail information, which deviated from our original intension of applying three-dimensional laser scanning technology in deformation monitoring.(3) We proposed to reconstruct surface using BP neural network, and reconstructed surface of the point clouds data before and after deformation in Matlab. This method not only can be drawn the entire region subsidence, but can also get the sinking value of any point. The fitting accuracy of this method is generally mm level; the greater the amount of point clouds data, the higher the accuracy, but almost no effect on computation.