Mems Laser 3D Vision System Image Processing

The laser 3D vision system has the advantages of high spatial resolution and no interference from ground clutter,even in the night and complex environment can also get three-dimensional information of spatial objects.So it has important application values in unmanned vehicles,intelligent robots and other environmental detection areas.At present,the mainstream laser 3D vision system generally obtains two-dimensional coordinate information of the space objects through the scanning system,obtains onedimensional longitudinal distance information of the spatial objects through the laser ranging system,and then combines two-dimensional information with the onedimensional distance information,at last forms three-dimensional information of spatial objects.In this paper,we focus on the key problems of spatial information modeling,laser point cloud data processing and 3D reconstruction of laser point cloud in 3D laser vision system.Finally,the experimental data obtained by MEMS laser 3D vision system are used to verify the result.The main contents are as follow:(1)The working principle of MEMS laser 3D vision system is briefly introduced.The relationship between the MEMS scanning angle and the control voltage was measured experimentally to find the MEMS scanning trajectory with high spatial coverage.The radial distance variation model between the laser 3D vision system and the target is established.Three-dimensional image distortion of different viewing field and different detection distance are simulated,and image correction methods are researched.The transformation model from geodetic coordinate system to laser 3D vision system coordinate system is established,which lays the theoretical foundation for positioning with GPS or Beidou navigation system.(2)The pre-processing method of laser point cloud data is researched.A comparative analysis is done between various point cloud neighborhood search algorithm,then we select kd-tree as the point cloud data search algorithm and perform the normal vector estimation.The singularities are eliminated by statistical filtering method.The method of smooth filtering is compared,and the method of bilateral filtering is optimized by simulation.(3)The algorithm of laser point cloud 3D image reconstruction is studied.According to the principle of Delaunay triangle reconstruction and the principle of triangular mesh optimization,combining the characteristics of MEMS scan,the direct triangulation method based on triangular mesh growth method is adopted.And the filtered frustum of a cone is simulated.(4)The existing MEMS scanning laser 3D vision experiment system is used to image the short range single target,and an effective calibration scheme is proposed in the process of scanning(x,y)coordinates and z coordinates without one-to-one correspondence.The complex targets are imaged at different distances,and the laser cloud data are preprocessed and reconstructed.The feasibility of the algorithm is verified.