一种改进ORB-SLAM的变电设备在线三维建模算法

An Improvement of Monocular ORB-SLAM Based on GMS Matching Algorithm

针对传统的人工三维建模费时费力的问题,引入ORB-SLAM算法实现在线三维建模。ORB-SLAM算法通过对单目相机提取的图片进行特征提取,可实现低成本实时生成三维点云,即三维模型,但算法本身依赖于视觉词典的建立。实际应用中,受条件的限制,往往视觉词典难以制作。通过引入GMS匹配算法替换ORB-SLAM原算法中相应模块,在不需要词典向量的条件下成功实现了特征匹配,并利用KITTI数据集对改进算法进行了仿真实验。结果表明,该改进算法能保证原有建图精度和运行速度。最后用改进算法对变电站设备成功进行了在线三维建模。

To solve the problem that traditional artificial 3D modeling of substations waste time and energy,ORBSLAM algorithm was introduced to achieve 3D modeling online.ORB-SLAM extracts features on images from monocular camera,and gives 3D pointcloud which is the 3D model cheaply at real time.However,ORB-SLAM needs a visual dictionary.The visual dictionary needs lots of images from the ORB-SLAM application scene,then extracts features from all the images and deal data,which wastes lots of time.Environment of substations is special,which has strict personnel access restrictions,and they lie in different areas,having different devices,so it’s very difficult to get enough images and deal them ahead of time.This paper used GMS matching algorithm to get place of ORB-SLAM matching section,so features can be matched without of dictionary vector.The improved algorithm was verified with KITTI dataset,and the results show that it will not decrease the accuracy and speed.Finally the improved algorithm was used to get 3D model successfully for substation equipment.