Research On 3D Imaging Method Of Near Field Target Based On MEMS Microphone Array

The mobile robot needs to perform scene perception when working in an unknown environment.Because traditional imaging equipment such as optics and radar can not work well under certain special environments like nuclear radiation and darkness,the solution of three-dimensional target imaging by using ultrasonic echo signals under near-field condition is proposed.The proposed method is discussed in detail through three aspects,which are the analysis and design of microphone receiving array,the solution of compressed sensing to near-field observation model and the denoising of point cloud including three-dimensional reconstruction.And finally the simulation verification of ultrasonic three-dimensional imaging was performed.The received signals of the array are derived under different sound fields,sound source signals and array structures.The traditional MUSIC algorithm is used to estimate the spatial orientation of the sound source points in the space,and the simulation analysis is performed under different array element spacings and array elements of the planar array.Based on the sound field characteristics of the ultrasonic transmitting array elements and the size of MEMS microphone receiving array elements,the rectangular array designed to use in the subsequent near-field observation model is optimized.By dividing the reconstruction plane of near-field object,it is assumed that the echo signals reflection points are on the grid points where the reconstruction plane intersects the surface of the object,and the reflection points are gained by the received signals of the microphone array.Under the near-field conditions,the traditional direction of arrival estimation require a priori information of the number of sound sources and have the disadvantage of low resolution,considering the sparseness of spatial reflection points,the theory of compressed sensing is used to construct the l0 norm and l1 norm sparse reconstruction optimization problems.A suitable measurement matrix was constructed,and different greedy algorithms were used to reconstruct the signal while the signal sparsity is known and unknown respectively.The simulation experiment proves that when the reconstruction signal sparsity is less than one fifth of the number of receiving array elements,the signal can be completely reconstructed.The signals reconstructed by the orthogonal matching pursuit algorithm and the piecewise orthogonal matching pursuit algorithm are basically the same.As the object scattered point cloud formed by the reflection points estimated by the sparse reconstruction algorithm from multi-angle scanning signals contains noise points,it is proposed to set the Gaussian statistical threshold to remove the outliers and guide filtering to smooth the local noise points after setting the KD tree index structure to the point cloud.The guided filtering is compared with the mean filtering and the bilateral filtering algorithm.The simulation experiment proves that the guiding filtering has the advantages of low algorithm time consumption and good local geometric feature retention.Since the pointcloud maybe sparse and can not well reflect the shape information of the space object,the Delaunay space mesh generation method is proposed to reconstruct the point cloud triangle mesh.The ultrasonic emission signal is acquired by an oscilloscope,the signal expression is obtained by MATLAB sampling and fitting.The proposed three-dimensional imaging method is verified by setting the same point source signal as the reflection point on the surface of the cylinder model as the reflection point.