Research On 3D Reconstruction Technology Based On Laser Vision Fusion

Three-dimensional technology is widely used in augmented reality,medicine,smart city,cultural relics protection,industry and other fields.Lidar and camera are commonly used sensors in 3D reconstruction.Lidar ranging is accurate,but lacks environmental texture information.Visual images contain rich semantic information,but are susceptible to environmental factors.The fusion of laser and vision can make up for the defects of a single sensor,adapt to complex environmental changes and obtain 3D scene data with rich information.In this paper,the 3D reconstruction technology of laser and vision fusion is studied for indoor environment.The laser point cloud registration method,three-dimensional reconstruction method and laser and vision fusion method are studied.Using ROS,PCL,Open CV and other software development environments,a 3D reconstruction system research platform was built.Aiming at the problems of large error,long time-consuming and slow convergence of traditional ICP algorithm,a point cloud registration method based on curvature feature constraint 3D-Harris algorithm combined with 3D shape context(3DSC)feature is proposed.On the basis of extracting the voxel center of the point cloud data,the k-nearest neighbor search is used to obtain the voxel center adjacent point as the voxel grid for down-sampling,which improves the rapidity of the algorithm.Taking the surface normal and mean curvature as the feature constraints,the 3D-Harris feature points after down-sampling are extracted and the feature points are described by 3DSC to improve the registration accuracy.Aiming at the problems of time-consuming,rough surface and holes in greedy projection triangulation algorithm,an improved method is proposed.Firstly,the voxel filter is improved,the optimal voxel grid is calculated with the number of point clouds as the threshold,and the center of gravity adjacent points are used instead of voxels to realize down-sampling,which improves the efficiency of the algorithm.Then the octree is used instead of KD-Tree to search the neighborhood point cloud,and the moving least square method optimized by octree is used to evaluate the normal direction,which shortens the backtracking time and improves the smoothness of the reconstructed surface.Aiming at the problem of sparse laser reconstruction data and lack of local details in indoor environment,a reconstruction method of laser point cloud and visual data fusion is studied.The laser pose is used as the predicted value,and the visual pose is used as the measured value.The EKF is used to fuse the laser and vision.In order to reduce the cumulative error,the loop closure detection combining laser and vision is adopted.Firstly,the initial estimation is carried out based on the visual word bag,and then the initial estimation is optimized by laser loop closure.Experiments show that the reconstruction effect of multi-sensor fusion is higher than that of single sensor in local fineness and integrity.