摘要
针对基于视觉的3D场景重建领域中设备昂贵、精度低、过程耗时的问题,提出一种采用RealSense摄像头,以OpenVINO构建底层计算平台,实现室内3D场景重建的方法。首先在数据获取部分,通过RealSense深度相机获取数据,实现了RGB数据、深度数据与点云数据的同时获取;其次,在计算加速部分,采用以OpenVINO为开发工具的神经计算棒构建硬件计算平台,提高了数据的处理速度,减少了计算开销;最后在点云数据的配准与融合部分,基于传统ICP算法进行了改进,大幅度减少了所需计算的数据量,提高了算法效率与精确度。
In allusion to the problems of expensive equipments,low precision and time-consuming process in the field of vision-based 3D scene reconstruction, a method of realizing indoor 3D scene reconstruction by building the underlayer computing platform with OpenVINO and based on RealSense camera is proposed. In the data acquisition part,the RealSense depth camera is used to obtain data to realize the simultaneous acquisition of RGB data,depth data and point cloud data. In the calculation acceleration part,the neural computing rod with OpenVINO as the development tool is used to construct the hardware computing platform,which improves the processing speed of data and reduces the computational overhead. In the registration and fusion part of point cloud data,the algorithm is improved based on the traditional ICP algorithm,which greatly reduces the amount of data required for calculation and improves the efficiency and accuracy of the algorithm.
作者
宋文龙
赵永辉
刘孟祎
SONG Wenlong;ZHAO Yonghui;LIU Mengyi(Northeast Forestry University,Harbin 150040,China)
出处
《现代电子技术》
北大核心
2020年第8期161-165,共5页
Modern Electronics Technique
基金
国家自然科学基金资助项目(31470714)。
