基于图像复原和亮度修正的口罩佩戴检测方法研究  

Research on Mask Wearing Detection Method Based on Deblurring and Light Compensation

作　　者：管晨曦 王淑营[1] GUAN Chenxi;WANG Shuying(School of Tangshan Graduate,Southwest Jiaotong University,Tangshan 063000)

机构地区：[1]西南交通大学唐山研究生院,唐山063000

出　　处：《计算机与数字工程》2023年第11期2671-2676,共6页Computer & Digital Engineering

摘　　要：在火车站等公共场合采集乘客图像时,由于运动模糊和光线问题,导致图像的质量受影响。为了提高检测口罩检测的准确率,从图像的复原和亮度修正两方面进行改进。首先,采用改进的维纳滤波算法对火车站乘客的模糊图像进行处理和复原。其次,将图像进行两次颜色空间转换,同时使用同态滤波算法对亮度进行调整,解决图像存在的光线过亮或过暗的问题。最后,对YOLOv4的网络参数进行调整,使用YOLOv4算法分别对原始的数据集和去模糊、修正亮度后的数据集进行训练和测试。试验结果表明,经过图像复原和亮度修正的数据集,相比于原始数据集,训练出来的模型更为优秀,平均检测精度最高提升了7.25%。不仅提高了口罩佩戴检测的准确率,而且满足了实时检测的需要。When collecting passenger images in public places such as railway stations,the image quality is affected due to mo⁃tion blur and light problems.In order to improve the accuracy of mask detection,image restoration and brightness correction are im⁃proved.Firstly,the improved Wiener filtering algorithm is used to process and restore the blurred image of railway station passen⁃gers.Secondly,the image is transformed into color space twice,and the homomorphic filtering algorithm is used to adjust the bright⁃ness to solve the problem of too bright or too dark light in the image.Finally,the network parameters of YOLOv4 are adjusted,and the original data set and the data set after deblurring and brightness correction are trained and tested by using YOLOv4 algorithm.The experimental results show that the trained model is better than the original data set,and the average detection accuracy is im⁃proved by 7.25%.It not only improves the accuracy of mask wearing detection,but also meets the needs of real-time detection.

关 键 词：口罩佩戴检测 图像复原 亮度补偿 YOLOv4 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]