基于图像光流的轴承滚子表面缺陷检测

Surface defect detection of bearing roller based on image optical flow

针对轴承滚子表面缺陷人工检测效率低、误检率高的问题,以及现有机器视觉检测方法存在的不足。基于上述情况,提出了一种基于图像光流的轴承滚子表面缺陷的检测方法。首先设计了一套图像采集装置,用于获取轴承滚子表面图像;其次采用全局和局部相结合的平滑策略,建立光流误差估计模型,同时为了增强缺陷位置的光流边界,在此模型中引入各项异性的扩散张量;在光流求解策略上,先对原始图像进行奇异值分解,消除灰度值异常对光流计算的影响,再引入金字塔分层细化方法,提高光流计算的准确性;最后根据光流生成的伪彩色图像,完成对缺陷区域的粗略定位,分割出缺陷区域。实验表明,所提方法在轴承滚子表面缺陷检测方面的可行性和有效性。

In the domain of bearing roller surface defect detection,both the human visual inspection and machine vision have problems,including low efficiency,high error rate,etc. To solve these problems,a detection method of bearing roller surface defects based on image optical flow is proposed in this article. First,a set of image acquisition devices is designed to obtain the image of bearing roller surface. Secondly,the combination of global and local smoothing strategy is adopted to formulate the estimation error model. At the same time,to enhance the optical flow boundary of the defect position,the diffusion tensor of the anisotropy is introduced in the model. For the solution strategy of optical flow,this paper carries out the singular value decomposition on the original image to eliminate the influence of gray value anomalies. The pyramid stratification method is used to improve the accuracy of optical flow calculations. Finally,according to the pseudo-color image generated by the optical flow,a rough location of the defect area is completed to segment the defect area. Experimental results show that the proposed method is feasible and effective on detecting the bearing roller surface defect.