由人工在生产线上进行汽车表面涂装的瑕疵检测和分类存在效率低、人工成本高等缺点。随着汽车生产自动化程度的提高, 迫切需要对检测过程实现自动化和智能化, 提出一种基于机器视觉方法, 利用图像纹理特征实现对汽车涂装瑕疵检测和分类。首先采用提取瑕疵局部二值模式(LBP)特征谱对汽车表面的喷涂瑕疵进行检测, 以克服背景光照和方向性对瑕疵特征检测的影响, 再利用Adaboost算法对选取的样本进行训练, 得到瑕疵种类分类器, 完成对检测结果的分类。通过实验所提出的方法, 对所选的几种典型瑕疵进行检测的有效率达到91%以上, 分类正确率平均达到82%以上, 单张图片检测时间控制在60 ms以下。

On-line detectionand classification for defect of automobile surface coating by manual inspectionhave many disadvantages, such as low efficiency, higher laborcost, etc. As automobile industry grow more autonomous, it is urgent that the coating defect detection and classification can be finished automatically and intelligently. This paper proposed a machine vision method to detect and classify the coating defects by texture analysis. Firstly,in order to overcome the effect of illumination and direction,the Local Binary Pattern (LBP) features is used to detect and extract the coating defects, and then the classifiers specific to some certain kinds of defects are obtained by training the defect samples using Adaboost algorithm, which can realized classificationfor all detected results. Experiment results show that the effective rate of this method can reach 91 percent and the average classification accuracy is higher than 82 percent, the detection time of a single picture is less than 60ms.