Research On Algorithm Of Poing Cloud Data Preprocessing Based On Two Oscillating Mirrors

With the invention and rapid development of 3D laser scanning as well as the processing methods, point cloud data has become a novel way of modeling objects with its property of easy-get. And this kind of new media has been widely used. The structure of two oscillating mirrors, a type of point cloud acquisition platform, has already become popular due to the platform’s easy to implement. Moreover, two-oscillating mirrors suits for a long-distance and large-size scanning with high scanning speed. However, this kind of structure can easily been disturbed and reduce a low precision of point clouds. Thus there is a need of preprocessing, especially a de-noising of point cloud.The algorithm of preprocessing can be implemented either in point model or in polygon mesh model. With the advantage of flexibility and high precision, the all algorithms of preprocessing and denoising almost implement in polygon mesh. This thesis mainly describes the designing of bilateral filter based on facet norm, the designing of two kinds of discrete differential edge operators; one is based on the outer norm of edge and another is based on the area of triangular polygon mesh. There is also an optimization of the function that denoises the 3D model. The main content of this thesis is as fellows.The first part of this thesis is to describe the basic knowledge of preprocessing algorithm of point clouds, followed by the description of denoising algorithm, which plays an important role in this thesis. With regard to denoising, this thesis firstly analysis the current algorithm and do some modification, then propose a new bilateral filter based on facet norm. This algorithm achieves an impressing effect of denoising. But considering the practical usage, the newly proposed algorithm is not suitable for extensive scanning.Thus this thesis then design a facet norm based discrete differential operator for triangular polygon mesh denoising. The divergence theory is firstly applied to the discrete edge to construct the operator. Due to the degradation of triangular, a numerical problem may occur during the iterations. Thus, an optimization of the operator is needed and this thesis replace the problematic operator with a new discrete operator based on the area of triangular polygon mesh. The newly proposed operator has the property of anti-degeneration of mesh model. This thesis use the later operator to build the target function of removing noises.Due to the nature of L0 norm exiting in the function, it becomes a NP-hard problem to solve the target function. By introducing a new parameter δ, the target function can been split into a half-quadratic minimization. Its huge benefit is that the difficult-solved function can be split into two easy-solved ones. A controlled iteration can act and an approximate solution is produced at the end of iteration.Our algorithm of denoising has high performance. Both little object with high resolution and large and distant object with sparse point cloud are processing effectively.