At the same time, the inconsistency of spatial coordinate systems further exacerbates the complexity of recognition problems, making traditional recognition methods face problems such as inaccurate feature extraction, low matching efficiency, and low recognition rate. Therefore, a 3D laser point cloud recognition method based on point-to-point feature algorithm and SVD decomposition is proposed. Firstly, the ISS key feature points are extracted from the point cloud data, and through the SVD decomposition algorithm, the identified ISS key feature point cloud and database reference point cloud are adjusted to a unified spatial coordinate system. By combining the FPFH descriptor with the SPFH graph, the spatial geometric characteristics of key point cloud feature points are described, and the point cloud data in the reference point cloud that meets the conditions of closest spatial proximity and the most similar FPFH descriptor is divided into a pair of feature point pairs. Introducing spherical harmonic function and calculating the similarity of point to point features to achieve the recognition of three-dimensional point clouds. The experiment shows that the proposed method can simplify complex point cloud data while preserving local ISS key feature points of the point cloud. Through feature point pair similarity analysis, effective recognition of 3D point cloud data can be achieved.
将通过ISS特征点选择后的点云数据作为待识别点云数据集,并将其转换到与数据库中参考点云相同的坐标系,明确数据库内点云数据的三维空间后,将待识别点云数据集视为源点云,并用表示,对源点云实施空间变换,使其与数据库中的参考点云在三维空间中的位置和方向一致,设经空间变换后得到的理想点云数据集为,利用奇异值分解(Singular value decomposition,SVD)算法[8],对到的变换过程展开求解,使其与数据库中的参考点云在三维空间中的位置和方向一致。SVD算法是一种矩阵分解方法,可以将一个矩阵分解为3个矩阵的乘积,帮助求解点云数据集间的最佳旋转和平移关系,从而实现点云的配准和匹配,进而实现点云的有效识别。
点对匹配则是建立在点云配准的基础上,通过寻找点云之间的精确对应关系来进一步提高识别的准确性。因此,利用快速点特征直方图(Fast point feature histograms,FPFH)描述子[9]对待识别点云内的特征信息实施描述。对待识别点云中的每个ISS特征点,计算其FPFH描述子,并与参考点云中的对应点进行比较,以找到最相似的匹配对。设待识别点云集内的随机ISS特征点为,其邻域特征点为,两特征点的法向量分别为,建立局部三维坐标系,具体的构建过程如下:
ZuoNi-na, QinXiao. Simulation of point cloud data extraction from 3d virtual image of object surface[J]. Computer Simulation, 2023,40(1):255-258.
[5]
HuiL, ChengM, XieJ, et al. Efficient 3D point cloud feature learning for large-scale place recognition[J]. IEEE Transactions on Image Processing, 2022, 31: 1258-1270.
LiKe. 3D point cloud research on an identification method based on PSO-KNN substation equipment[J]. Power System Protection and Control, 2021,49(18): 182-187.
YangJun, WangLian-jia. Recognition and segmentation of 3d point cloud through positional relation convolution in combination with deep residual network[J]. Journal of Xi'an Jiaotong University, 2023,57(5): 182-193.
LuoYong, ChangJing, YuanQian-jin, et al. A fast 3d point cloud recognition method for substation equipment[J]. Journal of Zhengzhou University (Engineering Science), 2023,44(3): 62-68.
HuangSi-yi, ChenXing-rong. Potential semantic analysis based on power iteration random singular value decomposition[J]. Journal of Xiamen University (Natural Science Edition), 2023,62(4): 679-686.
JiangFeng-yang, LiuYong-gang, ChenZhi-hang, et al. Vehicle mounted LiDAR point cloud registration method based on improved FPFH-ICP[J]. Journal of Chongqing University, 2023,46(5): 1-10.
WangLi-qun, LuXin, ShiLi-wei. Single-mode method for calculating the second partial derivatives of a single eigentriple in a quadratic eigenvalue problem[J]. Acta Mathematica Sinica (Chinese Series), 2022, 65(3): 559-570.
LiPeng, MaHao-yu, TianBing, et al. Current inversion method for magnetic field sensing data based on spherical harmonic function decomposition in complex electromagnetic environments[J]. High Voltage Engineering, 2023, 49(6): 2365-2373.