基于深度学习的双目视觉汽车压盘装配生产线研究

Research on Assembly Line of Binocular Vision Automobile Pressure Plate Based on Deep Learning

汽车工业自动化程度的提高致使越来越多的智能化设备代替人力资源,其自动化生产线中采用工业机器人视觉检测代替人工检测已经成为一个趋势,大大降低了人工劳动力的成本,但也存在分辩能力较差,缺乏适应性和智能化等问题。为了解决这一难题,提出了一种改进型深度学习算法用于工业机器人汽车压盘装配生产线工件的检测。算法研究将汽车压盘分解为4个单个工件,采用双目视觉技术提取采集立体图像,运用ReliefF算法提取工件的最优缺陷特征,运用“缩小型”CNN深度学习算法进行立体图像匹配。通过在不同环境下进行了多次实验,表明该算法可以对不同种类、材料的工件表面不同等级的缺陷进行精确、实时的深度学习并实现检测。

Automotive industry improvement leads to more and more automation intelligent equipment instead of human resources.Using industrial robot visual inspection in the automatic production line to replace manual detection has become a trend,greatly reduces the manual labor costs,but there are also problems such as poor resolution,lack of adaptability and intelligence.In order to solve this problem,this thesis introduces an Improved deep learning algorithm used in the detection of industrial robot car pressure plate assembly line workpiece.In the the algorithm the car pressure plate is decomposed into four individual artifacts,and the binocular stereo vision technology is used to extracted stereo image,ReliefF algorithm to extract the optimal defect characteristics of workpiece."Minification"CNN deep learning algorithm has been used to carry out stereo image matching,and many experiments were carried out in different environments to carry out accurate and real-time deep learning and detection of defects at different levels on the surface of workpiece of diff erent types and materials.