Research On License Plate Recognition Algorithm In Haze Weather

After the outbreak of the epidemic,in order to isolate the source of infection and prevent the spread of the epidemic,it is required in many regions to wear masks,monitor temperature,and scan code to display before entering and leaving public places.Due to the impact of multiple populations and small targets in complex environments on mask wearing detection,efficient contactless and intelligent wear detection is of great significance.In order to solve the above problems,this topic develops a deep learning algorithm and builds a mask detection system to achieve contactless intelligent detection of whether a mask is worn.The main work of this article is as follows:(1)Download the online open source dataset and two parts,personal photography and crawler mask images,to form this dataset.The dataset includes two categories:photos without masks and photos with masks,taken from different scenes,different lighting,different angles,and photos with masks of different styles and colors,totaling9800 images,and perform frequency domain image enhancement algorithms,Mosaic Mix up image fusion algorithm and other preprocessing methods expand the data volume,and then label the dataset through Label Img software,and obtain 12 sets of prior prediction frames through clustering algorithm.(2)A detection algorithm for mask wearing in complex environments is proposed to address the detection challenges of multiple crowds and small targets under multiple interference factors such as different environments and light source angles during mask wearing recognition.Firstly,the CSP module was changed to CSP1-1,and an attention mechanism module was introduced to reduce residual components and operations in the model,while also increasing the attention of mask features.Secondly,to solve the problem of small mask targets in the far field of vision,effective feature layers and network prediction scales are added in the SPP module to further improve the ability of extracting features from different Receptive field.The improved algorithm through experiments achieves an average accuracy of 96.6% for both non wearing and wearing masks.The FPS size of a single detection photo is 47.8,which increases m AP by 0.6%and FPS by 30 compared to the YOLOV5 algorithm.This results in good detection accuracy and speed for mask wearing,with real-time detection capability.(3)Design a portable and stable mask recognition system based on Linux system.The system is mainly divided into two parts to build: hardware environment and software environment.In order to expand to meet different requirements,a GPIO driver based on Linux system is designed,and the program is ported to Firefly through the conversion of the pt-onnx-rknn model＿The RK3399 pro development board,coupled with HIKVISION-E14 S high-definition 2K cameras and other external devices,completes the implementation of the mask wearing system,making the entire mask detection system more portable and universally adaptable.(4)Based on tools such as Py Qt Designer and other auxiliary packages,the design of the application software for the mask wearing detection system was completed.The overall architecture design of the software system has been completed through three layers: the interface layer,the logic layer,and the data layer.The Python files of the application software have been transplanted to the development board and combined with model program calls,realizing the functions of the entire system for image,video,and real-time detection of mask wear.The improved YOLOv5 mask wearing detection algorithm studied in this article is capable of completing mask wearing detection tasks in complex scenarios,while also being ported to firefly＿The RK3399 pro development board not only meets the detection requirements for mask detection,but also reduces hardware costs and unnecessary waste,and has good application prospects.