Research On Character Recognition Methods For Automobile Tires And Design Of Grouping System

The characters on automotive tires are imprinted by tire manufacturers on the side surface of the tire during production to record parameter information,including place of origin,factory code,and production date.To ensure that tires being sold to the market are not stored for too long,a manual character recognition method is used to select tires that have been stored beyond a certain period before they are shipped.Exceeding a certain storage period can cause rubber aging,which affects usage safety and can result in huge losses for customers and manufacturers.If automatic identification technology is used to identify tire characters and group overdue tires,it will not only reduce the consumption of manpower and material resources,but also realize the traceability and management of tires by manufacturers.In response to the above problems,this thesis designs a car tire character recognition and grouping system.The main work focuses on the following aspects:Firstly,considering the factory warehouse environment and tire imaging characteristics,the entire system is structurally designed in this study.Suitable cameras,lenses,and light sources are selected to establish an image acquisition device for real-time capture of automotive tires.The captured images undergo preprocessing operations to enhance the contrast between the tire character regions and the background regions.Additionally,a polar coordinate transformation is applied to convert the tire’s circular region into a rectangular region.Subsequently,tire character localization is performed.Due to the fixed relative position between the DOT identification characters and the characters to be recognized,the position information of the recognized characters is determined by locating the DOT characters.Template matching,Faster R-CNN,Yolov4,and an improved version of Yolov4 are employed for comparative experiments in character localization.Experimental results demonstrate that replacing the backbone network of Yolov4 and introducing attention mechanisms lead to significant improvements in localization speed,achieving a frame rate of 101.4789 frames per second and an accuracy of 97.76%.These findings indicate that the improved Yolov4 approach exhibits excellent performance in tire character localization in this study.Subsequently,the recognized tire characters after localization are subjected to character recognition.Multiple features are extracted from the character images and fed into a support vector machine(SVM)for model training.In order to improve upon the classical LeNet-5 network,modifications such as introducing batch normalization(BN)layers and incorporating Inception modules are proposed.Recognition experiments are conducted using SVM,LeNet-5 network,and the improved LeNet-5 network.Ultimately,it is found that the improved LeNet-5 network achieves an enhanced recognition accuracy of99.4%,which is 3.3% higher compared to the classical LeNet-5 network.The recognition speed reaches 64 milliseconds per image.Finally,the system software design is implemented on the Qt software platform.The system loads pre-trained models and enables online recognition and detection.It provides the capability to directly view recognition results and grouping results on the main interface.Furthermore,it allows uploading recognition data to a database,facilitating remote operation and viewing for convenient tire traceability management.