Assembling crankshaft bearing caps (CBCs) on automobile engine is a vital and complicated task that manual work is usually inefficient and become more prone to error. In this article, an autofeeding system with vision and line laser combined with an industrial object location and classification method is proposed for assembling the CBCs. First, an improved fully convolution one-stage network is used to obtain 2-D localization and detect the placement direction and order of a group of CBCs in the CBCs' image. Based on the 2-D location information, the line laser projects the laser on the surface of the CBCs at two different positions, and the images are captured accordingly. Then, an improved Steger algorithm is proposed to extract the centerline of the laser light bar, and the two laser images are merged into one laser image to calculate the height and pose information of the CBCs. The experimental results demonstrate that our method has achieved high real-time performance, accuracy, and robustness. After comparison and application in the factory, it is illustrated that the proposed autofeeding system improves the productivity and saving manpower.

中文翻译：

---

基于3D视觉引导的汽车发动机CBC自动上料系统


在汽车发动机上组装曲轴轴承盖 (CBC) 是一项至关重要且复杂的任务，手工作业通常效率低下且更容易出错。在本文中，提出了一种采用视觉和线激光并结合工业对象定位和分类方法的自动送料系统来组装 CBC。首先，使用改进的全卷积单级网络来获得二维定位并检测CBC图像中一组CBC的放置方向和顺序。根据二维位置信息，线激光将激光投射到 CBC 表面的两个不同位置，并相应地捕获图像。然后，提出一种改进的Steger算法来提取激光光条的中心线，并将两幅激光图像合并为一幅激光图像，以计算CBC的高度和位姿信息。实验结果表明，我们的方法具有较高的实时性、准确性和鲁棒性。经过工厂的对比和应用，表明所提出的自动送料系统提高了生产率并节省了人力。