3D-SIFT特征提取与体素滤波结合的点云精简算法

Point cloud reduction algorithm of 3D-SIFT feature point extraction and voxel filtering

点云数据中的冗余数据会影响到点云处理算法的速度,因此,为提升算法速率,需对点云数量进行精简。然而,点云精简过程容易剔除掉特征点,导致点云信息不完整,效果不好等问题。针对这些问题,提出一种利用3D-SIFT特征提取与八叉树体素滤波结合的点云精简方法。利用3D-SIFT算法提取出点云的强特征点和弱特征点,对弱特征点进行改进的八叉树体素滤波,并保留强特征点,通过点云合并,将滤波后的弱特征点与保留的特征点整合到一起,使得精简后的点云数据不丢失特征点信息,从而也达到了精简的效果。将本算法与均匀网格算法、非均匀网格法、随机采样算法进行对比实验。通过多个不同模型的可视化结果和信息熵评价分析,可以得出对于几种不同模型取平均本算法平均信息熵达到3.771 92,高于其他算法的信息熵,证明本算法在对数据进行精简的同时也达到了特征保留的效果。

Redundant data in point cloud data will affect the speed of the point cloud processing algorithm.Therefore,simplify the number of point clouds to improve the algorithm speed.However,the process of point cloud reduction is easy to eliminate feature points,resulting in incomplete point cloud information and poor effect.To solve these problems,this paper proposes a point cloud simplification method combining 3D-SIFT feature extraction and octree voxel filtering.The strong feature points and weak feature points of point cloud were extracted by 3D-SIFT algorithm,the weak feature points were filtered by improved octree voxel filtering,and the strong feature points were retained,the filtered weak feature points were integrated with the retained feature points,so that the simplified point cloud data didn’t lose the feature point information,so as to achieve the effect of reduction.The proposed algorithm was compared with the uniform grid algorithm,non-uniform grid algorithm and random sampling algorithm.Through the visualization results and information entropy evaluation analysis of several different models,it could be concluded that the proposed algorithm achieves the effect of feature retention while simplifying data compared with other algorithms.