基于立体视觉的大尺寸工业测量系统

基于立体视觉的大尺寸工业测量系统 Large dimension industrial measurement system based on stereo vision 刘建伟 first-author 西安交通大学 机械工程学院,陕西 西安,710049 西安交通大学 机械工程学院,陕西 西安,710049 针对工业制造领域普遍存在的大型物体全尺寸测量难题，提出并实现了一种基于立体视觉技术的便携式工业测量系统。该系统采用多个CCD视觉传感器，获取被测物体的多视图像序列，通过前方交会由二维像点坐标计算物体表面点的三维坐标。全局关键点测量由多目立体视觉系统完成，使用CCD相机从多个位置和方向拍摄物体，获得多目视差照片，由测量软件二维处理并完成三维重建。局部区域密集点云的测量由双目立体觉系统完成，左右两相机同步拍摄由投影机投向物体表面的编码条纹，测量软件自动进行立体匹配和三维重建后获得物体表面局部的密集点云。最后测量软件根据局部的关键点完成密集点云的坐标系转换。实验表明，该系统的测量精度达到0.1mm/3m，可以满足工业现场大尺寸测量对精度和效率的要求。 In order to realize rapid measurement on large dimension object in industrial fields, a new measurement system based on Stereo Vision was proposed and implemented. The system employing several CCD as vision sensors obtains pictures of the object from deferent viewing positions, finds out 2D coordinates of the points in each pictures ,then calculates their 3D coordinates based on the method of intersection. Global reference points are inspected by Multi-View Stereo Vision system: a high-resolution CCD camera is used to take a group of pictures surrounding the object, after analyzing the obtained 2D parallax pictures, the measurement software carries out 3D reconstruction of the global references points. Local area dense points is inspected by Binocular Stereo System:, after focusing the measurement head to given area object, projecting the coded stripe on the surface, meanwhile capture pictures from the left and right CCD camera synchronously, the measurement software automatic carry out the stereo matching, reconstruction, result in local area dense points clouds. Also, software transforms the local dense points clouds to the global reference points coordinates system, according to the relation between local reference points and global reference points. Experiment show that the new system is capable of meeting precision and efficiency demand of large dimension measurement. 立体视觉 标志点识别 立体匹配 三维重建 点云配准 stereo vision Reference points detect stereo matching Reconstruction Points cloud alignment TP393 2009-01-08 2009-03-19 2010-01-22 2010-01-22 2020-08-11 1 18