基于机器视觉的汽车压装衬套偏转角度测量

Machine Vision-Based Measurement of Deflection Angle in Automotive Press-Fit Bushings

衬套安装的角度测量对于确保压装过程中汽车悬架与衬套的准确组装十分重要,现有的单目测量方法由于存在相机安装环境受限、测量准确率低的缺点而难以广泛应用。为提高衬套偏转角度的测量准确性,提出了一种基于特征匹配与径向基神经网络的衬套偏转角度测量方法。采用Hessian矩阵优化ORB算法,剔除误匹配对,提高ORB算法匹配性能;采用基准模板匹配策略,解决相机斜视状态下图像特征被遮挡导致的无法匹配问题,并将采集图像的特征点转换至基准模板上;通过引入径向基函数神经网络进行偏转角度软测量,拟合特征点与偏转角度的非线性关系,提高衬套偏转角度测量的精度。实验结果表明,所研究方法可以有效进行偏转角度测量,最大平均相对误差为2.72%,满足衬套偏转角度测量要求,在汽车生产过程中有一定的应用价值。

The angle measurement of trim panel installation is considered crucial for ensuring the accurate assembly of automotive suspension with the trim panel during the pressing process.Existing monocular measurement methods have been difficult to widely apply due to limitations imposed by the camera mounting environment and the low measurement accuracy.In order to enhance the measurement accuracy of trim panel rotation angle,a method based on feature matching and radial basis function neural network was proposed.The ORB algorithm was optimized using the Hessian matrix to eliminate mismatched pairs and improve the matching performance.A baseline template matching strategy was employed to address the issue of image feature occlusion caused by camera oblique view,and the feature points collected from images were transformed onto the baseline template.By introducing a radial basis function neural network for rotational angle soft measurement,the nonlinear relationship between feature points and rotation angle was fitted to enhance the accuracy of trim panel deflection angle measurement.Experimental results demonstrated that the proposed method effectively performed angle measurement,with a maximum average relative error of 2.72%,meeting the requirements of trim panel deflection angle measurement,and possessing certain application value in the automotive production process.