Research On Time-of-flight Subpixel Measurement Multi-source Data Fusion Algorithm

Three-dimensional imaging is a hotspot in the field of computational information technology,and it is widely used in many fields such as unmanned navigation,robotic navigation and deep space exploration.At present,the resolution of 3D imaging technology based on laser intensity modulation flight time is low.In this paper,subpixel data fusion algorithm is studied for 3D point cloud data.The specific research work of this paper includes the following aspects:The mathematical model of 3-D imaging and the theory of optical distortion correction are studied.The optical distortion of the time-of-flight sensor is corrected by using the standard plate.The geometric characteristics of the background,the plane features of the buildings and the geometric relations of the positions of the buildings in the indoor scene are studied.The K-nearest neighbor algorithm is studied and the outliers generated by noise in point cloud data are filtered by distance information.Experimental results show that the proposed algorithm is adaptable to the change of background geometry and the background of different materials,and can extract the point cloud effectively.The theoretical model of point cloud registration is studied,the registration method of point cloud data from different positions is studied,and the registration method for time-of-flight point cloud data is proposed.Aiming at the low efficiency of FPFH(Fast Point Feature Histogram)feature extraction coarse registration algorithm,KD(KDimensional)tree is used to optimize the efficiency of registration.Aiming at the low registration precision and efficiency of Iterative Closest Point algorithm,an improved Point-to-plane ICP algorithm based on normal vector angle sampling is proposed to realize fine registration.KD tree algorithm is used to optimize the search efficiency.At the same time,RANSAC(Random Sample Consensus)algorithm is used to remove the wrong corresponding points,and Point-to-plane matching method is used to reduce the registration error caused by the inconsistency between the two groups of point cloud data.Compared with other registration methods,the experimental results show that the proposed algorithm has higher efficiency,better stability,higher registration accuracy,and still has good registration effect.Through the statistical analysis of the data in the target region,the points after registration are optimized to the sub-pixel grid region,and the different point cloud data in each grid are optimized to the centroid of the sub-pixel grid.Then the point cloud surface is smoothed by K-nearest neighbor algorithm to reduce the fusion error caused by system error.The experimental results show that the average resolution of the target region is increased by 88.81%,the spatial resolution of the equivalent sensor is increased from 640 × 480 pixels to 860 × 640 pixels,the range resolution is improved by 15.51%,the structure of the point cloud is clearer and the geometric features are further enhanced.