基于透视形变校正与分层检测的机箱面板视觉检测方法

A Vision Inspection Method for Chassis Panels Based on Perspective Distortion Correction and Hierarchical Detection

摘要: 针对工业机箱面板自动化生产中的型号混料与表面缺陷在线检测难题，本文设计了一套基于机器视觉的在线检测系统。该系统旨在实现对10款不同型号机箱面板的准确区分，并完成对丝印缺失、丝印不同、I/O接口缺失及I/O接口不同共4类缺陷的可靠检测。方法上，创新性地采用了透视形变校正下的模板匹配与分层检测策略，首先通过大模板进行快速粗定位与姿态估计，进而映射出精确的检测区域，并利用小模板实现关键特征的精细匹配与缺陷判别。测试结果表明，系统在混料检测中型号区分准确率达100%；在实验室环境下，对多类缺陷的整体检出率为97.5%，误检率为0.5%；在产线连续运行测试中，系统综合准确率达99.70%，平均单件检测时间低于0.8秒，满足高速产线节拍要求。该系统在检测精度、效率与工程实用性方面表现优异，具备良好的抗透视形变能力与在线部署价值。

Abstract: To address the challenges of model mix-up and online surface defect detection in the automated production of industrial chassis panels, this paper presents a machine vision-based online inspection system. The system is designed to accurately distinguish between 10 different panel models and reliably detect four types of defects: missing silkscreen, inconsistent silkscreen, missing I/O interfaces, and inconsistent I/O interfaces. Methodologically, an innovative strategy integrating perspective distortion correction with template matching and hierarchical detection is proposed. Firstly, a large template is utilized for rapid coarse positioning and pose estimation, which subsequently enables the precise mapping of inspection regions. Then, small templates are employed for fine feature matching and defect discrimination. Test results demonstrate that the system achieves 100% accuracy in model differentiation during mix-up inspection. In laboratory tests, the overall defect detection rate reaches 97.5% with a false detection rate of 0.5%. In continuous production line testing, the system attains an integrated accuracy of 99.70% with an average processing time per unit below 0.8 seconds, meeting the requirements of high-speed production cycles. The system exhibits excellent performance in detection accuracy, efficiency, and engineering practicality, demonstrating strong robustness against perspective distortion and significant value for online deployment.