基于立体视觉的三维曲面测量关键技术研究

利用立体视觉测量曲面有三个难点一是摄像机标定,建立图像点与空间点的映射关系;二是立体匹配,在左右图像上寻找对应点;三是测量大型物体时,需要拼接图像.首先根据人眼感知事物的原理,利用神经网络拟合图像点与空间点的映射关系,避开直接对摄像机标定;接着以光栅投影曲线为特征,利用小波边缘检测和搜索式无监督聚类,结合视觉几何不变性,实现亚像素级的立体精匹配;最后采用小波多尺度多非分辨率的特性,拼接图像,融合数据,实现曲面全方位测量,且测量精度在0.5mm/m以内.

3D Surface Measurement Key Technique Based on Computer Vision

There are three difficulties in 3D measurement based on stereo vision: (1) Camera calibrate, establishing the mapping relation between image point and special point; (2) Stereo matching, searching corresponding point in a pair of stereo images; (3) Image mosaic when a large-scale object measured. In this paper, using neural network, the mapping relation between image points and special points is established. With wavelet edge detection, searching non-supervisor clustering and geometric invariance, distorted stripes that are modulated by surface shape of object when grating is projected on object are regarded as features to achieve stereo matching at subpixel level. Furthermore, the multi-scale and multi-resolution attributes of wavelet are applied to image mosaic and data integration, so a large scale object can be measured, and the measuring error is less than 0.5 mm/m.