基于计算机视觉的钢铁行业车牌识别算法

Computer Vision-based License Plate Recognition Algorithm for Steel Industry

针对重新训练模型数据收集成本较高、模型鲁棒性较低等难点,在PP-OCRv3模型的基础上,对后处理方法进行优化改进,并提出了适用于钢铁行业场景的车牌模型训练方法。在中国城市停车数据集及钢厂场景数据集中进行实验,实验结果证明提出的车牌识别算法部署在边缘设备上时能够保证推理时间小于200 ms的同时取得99.6%的识别精度,足以满足钢铁行业场景中车牌识别的精度和性能需求。

In view of these difficulties,which the self-made license plate in the steel industry cannot be recognized by the general license plate recognition algorithm,and the data collection cost of the retraining model is high and the robustness is low,the post-processing method is optimized and improved on the basis of the PP-OCRv3 model,and the model training method which is suitable for steel industry scenarios is proposed.Experiments are carried out on the CCPD and the steel factory scene datasets.The experimental results show that when deployed on edge devices,the inference time can be guaranteed to be less than 200 milliseconds and the recognition accuracy of 97.6%can be achieved,which is enough to meet the accuracy and performance requirements of license plate recognition in steel industry scenarios.