Research On Splicing Of Point Cloud And 3d Reconstruction Algorithm In Structured Light Measurement

With the development of autonomous driving technology,the acquisition of three-dimensional information about the road in front of an autonomous vehicle has become a necessary prerequisite for the safe driving of the vehicle.At this stage,the measurement of the road in front of the vehicle mainly focuses on the research of structured road surface,and there is less research on the measurement technology of unstructured road surface.For the unstructured road boundary is not obvious,the road surface is not flat and other characteristics,this thesis studies the point cloud stitching and three-dimensional reconstruction technology using the structured light method of road detection in front of the vehicle.According to the characteristics of unstructured roads,the principle of structured light measurement is analysed,and the general scheme of point cloud stitching and 3D reconstruction of unstructured road measurement system based on structured light is designed.The structured light spot extraction and point cloud data pre-processing techniques in the road surface image in front of the vehicle are studied.In the extraction of the structured light spot position in the image,firstly,the grey scale is greyed out;in order to eliminate the noise interference information in the image,the image is median filtered;and the Sobel edge detection algorithm is applied to detect the light spot position;for the light spot defects in the edge detection,the open and closed operations of the morphological processing method are applied to repair;finally,the grey scale centre of gravity method is applied to extract the light spot position Finally,the spot position is extracted using the grey centre of gravity method.The3 D sparse point cloud of the measured pavement is obtained by combining the structural light measurement principle and the structural parameters of the system to solve the spot coordinates.The Delaunay triangular segmentation algorithm is used to mesh the original 3D point cloud data in order to recover the surface of the road surface,and the LOOP segmentation and butterfly segmentation are applied to mesh the surface respectively.In order to obtain the complete point cloud data of the pavement,the point cloud data stitching technique of the sequence image is studied.The initial and target point clouds are solved for their rigid transformation matrices using the quaternion method.The Super-4pcs point cloud coarse stitching algorithm is applied to randomly select four points that are co-planar but not co-linear in the initial point cloud to form a quadrilateral,and use the affine invariance and the angle and distance constraints to filter the corresponding quadrilateral in the target point cloud to achieve the coarse stitching of the point cloud data.To address the errors arising from the coarse stitching,the ICP point cloud refinement stitching algorithm is improved,and the KD-tree is introduced to index the point clouds to speed up the search for corresponding points.The experimental data shows that the improved algorithm shortens the stitching time,reduces the stitching error,and improves the stitching accuracy and efficiency.For the discrete point cloud data of the road in front of the car,a Poisson equation-based surface reconstruction method is investigated for 3D reconstruction in order to avoid local segmentation of the surface.The indication function of the surface to be fitted is predicted,the equivalent surface is extracted based on this function,and the 3D surface model is finally fitted to the point cloud data.The experimental results show that the reconstructed 3D surface has good robustness and smoothness,and the reconstructed results are basically consistent with the original pavement structure,which can show the shape of the pavement in front of the car better.