针对逆向工程中引导性曲面边界信息的快速获取问题, 系统地研究了共轴立体视觉测量方法, 建立该方法的数学模型, 详细分析了摄像机焦距、基线距等系统结构参数及被测点空间位置对测量精度的影响, 通过数学分析确定摄像机基线距的最佳取值范围, 研究共轴立体视觉测量系统特殊的极线几何关系.提出基于共轴立体视觉的曲面边界快速测量方法, 利用三坐标测量机的精密机械系统及精确的空间定位能力, 用单个摄像机以两次共轴定位摄取图像的方式实现共轴立体视觉测量功能, 然后利用共轴立体视觉外极线相互平行且通过各自像平面主点的特殊极线几何关系简化同源像点匹配过程, 从而快速获取被测曲面的边界信息.实验结果表明: 用基于三坐标测量机的单摄像机共轴立体视觉测量方法获取的曲面边界平均误差为0.268 mm, 基本满足逆向工程中对引导性曲面边界的测量精度要求.

Low-level precision boundary information of the unknown surface is useful for digital sensor path planning, data segmentation and feature recognition in reverse engineering. In order to acquire the low-level boundary information of the measured surface both rapidly and easily, the Coaxial Stereo Vision Photogrammetry (CSVP) was systematically studied. The mathematical model of CSVP was established, the influences of system parameters including the focal length of the cameras, the baseline distance and the spatial position of the measured points on measuring accuracy were analyzed, and the mathematical analysis for determining the optimum range of the baseline distance of the two cameras was carried out. In addition, the characteristics of the CSVP epipolar geometry were studied. Then a new approach for rapid digitization of surface boundaries based on CSVP was proposed. This approach was characterized by an integrated use of a Coordinate Measuring Machine (CMM) and a single CCD camera, the CMM was employed to provide an accurate and repeatable platform for the camera. Consequently, the camera can be located at any predefined position within the CMM’s working volume. This enables the camera to get stereo image pairs in different positions along the same axis, which can only be achieved by two cameras without the CMM. During data processing, the characteristics of the CSVP epipolar geometry, in which the corresponding epipolar lines in the front and back image planes are arranged parallel to each other, and the epipolar lines pass through the main points of their own image planes respectively, were applied to facilitate stereo matching, by which the surface boundary information of the measured surface can be obtained both rapidly and easily compare to conventional binocular stereo vision method. Experimental results indicate that the average error of the surface boundaries reconstructed by the data gathered by the proposed method is 0.268 mm. It can meet the accuracy requirements of low-level boundary information in reverse engineering.