摘要
针对现代传感器由于自身扫描范围限制不能一次性完成对象的扫描,需重复多视角来获取对象完整三维信息的问题,提出一种通过传感器实时获取周围场景来实现三维点云拼接的方法。首先,传感器采集场景的深度信息和彩色信息来获取连续帧间的SURF特征点,并计算特征描述子;然后,采用局部特征描述的粗配准和ICP方法将不同视角下的点云转换到统一的坐标系下,并通过RANSAC算法剔除误匹配;最后,结合得到的平移矩阵和旋转矩阵完成点云数据的无缝拼接。实验结果表明,应用特征匹配方法进行点云拼接的算法在消除公共部分冗余数据的同时,能够比较完整的保留两组点云之间互补包含的点云数据。
For the modern sensors, due to the limitation of its own scanning range, it is not possible to complete the scanning of the complete model at a time. It is necessary to repeat the multi-angle of view to obtain the complete three-dimensional information of the model. The paper proposes a method for realizing the three-dimensional point cloud splicing by real-time acquisition of surrounding scenes through sensors. Firstly, the sensor collected the depth information and color information of the scene to obtain SURF feature points between successive frames, and calculateed the feature descriptors. Then, the coarse registration and ICP methods of local feature description were used to convert the point clouds to the united coordinate system, and the mismatch was eliminated by RANSAC algorithm. Finally, the translation matrix and rotation matrix were combined to complete the seamless mosaic of point cloud data. The experimental results show that the algorithm of point cloud splicing using feature matching method can completely preserve the redundant point data of the two parts of the point cloud while eliminating the redundant data of the common part.
作者
吕颖
陈志明
杨天天
牛康
LV Ying;CHEN Zhi-ming;YANG Tian-tian;NIU Kang(Nanjing University of Aeronautics&Astronautics,Nanjing Jiangsu 210016,China;Shanghai Electro-Mechanical Engineering Institute,Shanghai 201100,China)
出处
《计算机仿真》
北大核心
2020年第10期200-205,共6页
Computer Simulation
基金
国家自然科学基金(61673212)
江苏省自然基金(BK20161490)。
关键词
随机抽样一致性
特征点提取
点云特征匹配
点云拼接
RANSAC
Feature point extraction
Point cloud feature matching
Point cloud stitching
