基于单应性矩阵的大尺寸零件视觉测量图像拼接方法

摘要

针对视觉检测技术对大尺寸复杂零件测量精度低的问题,提出一种基于单应性矩阵的大尺寸零件视觉测量图像拼接方法;首先建立世界坐标系→相机坐标系→投影坐标系→图像坐标系之间的归一化模型,提炼出简单易懂的坐标变换表达式H;然后对摄像机进行多次标定,确定像素当量、拍摄视场及焦距;再获取零件图像,利用前面得到的最简坐标表达式H得到拼接图像之间的配准,最后完成若干副图像的拼接,把拼接结果和SURF特征点方法及向后映射模型归一化的拼接结果进行比较;结果表明,该方法可以大大减少大尺寸零件拼接的时间,较SURF特征点方法和向后映射拼接用时分别降低86.55％和67.30％,拼接精度远高于SURF特征点方法和向后映射方法,分别提高了50.95％和25.08％,测量精度满足大尺寸零件检测精度要求.%Based on the problem that the visual detection technology is low accuracy of detecting large-size and complex parts.A method that high precision and fast image mosaic method for large size component vision measurement based on homography transformation is proposed.Firstly the normalization model from the world coordinate system to the camera coordinate system,projection coordinates and image coordinates is established to extract the expression of simple coordinate transformation;Then the camera is calibrated to determine the pixel equivalent,the field of view and obtain the focal length;Images of component are obtained,and use the simplest coordinate expression according to the obtained image splicing between registration,and finally complete stitching of a number of images,compare this stitching results with the results of SURF and backward mapping model normalization.The result shows that this method can greatly reduce the mosaic time of the component,86.55％ and 67.30％ lower than the SURF and the backward mapping used for jointing,The precision is much higher than that of the backward mapping method and SURF,50.95 ％ and 25.08％.The measurement precision can meet the requirements of large size component.