基于自适应ROI的轴承内圈高精度视觉尺寸检测

High Precision Visual Dimensions Measurement of Bearing Inner Ring Based on Adaptive ROI

针对机器视觉轴承内圈侧面复杂形状尺寸检测精度低的问题,提出根据检测目标建立小面积感兴趣区域(Region of Interest,ROI)的自适应选取方法和基于Zernike矩的ROI亚像素级边缘提取方法,大幅提升了轴承内圈尺寸的检测精度。首先分别拍摄轴承内圈左侧与右侧轮廓图像,对图像进行预处理。在此基础上,通过角点检测融合像素扫描的方法实现自适应ROI选取,解决了因轴承内圈移动引起的小面积ROI边缘误判问题;使用Canny算子提取ROI的像素级边缘,再用改进的Zernike矩算法得到亚像素级边缘。最后,分别对ROI中提取的边缘进行最小二乘圆拟合和直线拟合,根据像素当量与视场间隔将图像中各尺寸转换为轴承内圈实际尺寸。实验结果表明:所提方法测量的标准不确定度低于0.005 mm,满足轴承尺寸高精度检测的要求,对于实现轴承检测的自动化有实际意义。

Targeting at the problem of low precision in the measurement for complex shape dimensions of the bearing inner ring side based on machine vision,the self-adaptive method for determining the small-area region of interest(ROI)according to the detection tar-get and the ROI sub-pixel edge extraction method based on Zernike moment are proposed,which greatly improved the measurement pre-cision of bearing inner ring dimensions.Firstly,the images of the left and right sides of the bearing inner ring are captured and prepro-cessed respectively.Then the adaptive ROI determination is achieved by corner detection and pixel scanning,which solve the misjudg-ment problem of small-area ROI edge caused by bearing inner ring movement.The pixel-level edges of the ROI are extracted by using the Canny operator,and then the sub-pixel-level edges are obtained by using an improved Zernike moment algorithm.Finally,the edges extracted from the ROI are fitted with the least-squares circle fitting and the linear fitting respectively,and each dimension in the image is converted to the actual size of the bearing inner ring according to the pixel equivalent and field-of-view interval.The experimental re-sults show that the standard uncertainty measured by the proposed approach is less than 0.005 mm,which meets the requirements of high-precision inspection of bearing dimensions and has practical significance for the automation of bearing inspection.