Abstract: Intelligent and automated industrial production is one of the foundations for improving the competitiveness of companies. By installing industrial cameras and light sources at workstations as the hardware system, and using Python as the programming language, and OpenCV and TensorFlow as the core to develop a machine vision based body vehicle identification number (VIN) recognition system as the software system , the body VIN images taken by the industrial cameras are trained and recognised for use to realise the intelligent operation of the body VIN scene in the field of vehicle manufacturing. The recognition accuracy rate currently reaches 99.8% after testing on site. In addition to the recognition of the images themselves, the system has also develope a unique four-fold error prevention strategy based on the process of the site to achieve the goal of "0" error outflow. The development of a machine vision based body VIN identification system has helped to improve the intelligence of industrial production and provided more possibilities for the development of self-sustainable systems for the use of machine vision in industrial production, improving the ability of enterprises to reduce costs and increase efficiency, and enhancing their competitiveness.

Key words: Machine vision; VIN code identification system; Vehicle manufacturing; Python; OpenCV; TensorFlow

摘要： 工业生产智能化、自动化是提高企业竞争力的基础之一。通过在工位安装工业相机及 光源作为硬件系统、以 Python 为编程语言、OpenCV 和 TensorFlow 为核心，开发基于机器视 觉的车身车辆识别代码（VIN）识别系统为软件系统，对工业相机拍摄的车身 VIN 码图片进 行训练和识别运用，实现整车制造领域车身 VIN 码场景的智能化作业的转变。经现场的测试 使用，识别准确率目前达 99.8%。在系统中除了图像的自身识别外，还根据现场的工艺制定 了独特四重防错策略，达到错误“0”流出的目标。通过开发基于机器视觉的车身 VIN 码识 别系统，对提高工业生产的智能化水平，并为机器视觉在工业生产中的运用提供了更多自立 化系统开发的可能，提高了企业降本增效的能力，提升企业的竞争力。

关键词: 机器视觉；VIN 识别系统；整车制造；Python；OpenCV；TensorFlow