基于3D-SIFT与SICP的激光点云与影像配准方法

Laser point cloud and image registration method based on 3D-SIFT and SICP

针对激光点云与影像配准数据量大、质量差的问题,本文提出一种基于三维尺度不变特征变换(3D-SIFT)与尺度迭代最近点算法(SICP)相结合的激光点云与影像配准方法。首先使用运动恢复结构(SfM)方法将影像通过光束法平差生成影像三维点云,然后使用3D-SIFT提取激光点云与影像三维点云中的特征点,接着利用对偶四元数求解激光点云和影像三维点云的初始变换矩阵,实现两种点云数据的粗配准;最后利用SICP算法实现两种点云数据的精配准。实验结果表明,本文方法获取的激光点云与影像配准中误差为0.24 cm,配准时间为69 s,且与选代最近点算法(ICP)相比提高了配准精度和配准效率。

Aiming at the problem of large amount of data and poor quality of laser point cloud and image registration,in this paper,a laser point cloud and image registration method based on 3D-scale Invariant feature transform(3D-SIFT)and scaling iterative closest point(SICP)was proposed.Firstly,the structure from motion(SfM)method was applied to generate the 3D point cloud of the image through the bundle adjustment.Then,the 3D-SIFT was explored to extract the feature points in the laser point cloud and the 3D point cloud of the image.Then,the dual quaternion was introduced to solve the initial transformation matrix of the laser point cloud and the 3D point cloud of the image,and the coarse registration of the two point cloud data was realized.Finally,the SICP algorithm was applied to achieve the fine alignment of two point cloud data.The experimental results showed that the error of laser point cloud and image registration obtained by this method was 0.24 cm,and the registration time was 69s,which improved the registration accuracy and efficiency compared with the ICP algorithm.