基于改进卷积神经网络的安全带佩戴识别  

Seat belt-wearing recognition based on improved convolutional neural network

作　　者：丁家益 周跃进 DING Jiayi;ZHOU Yuejin(School of Mathematics and Big Data,Anhui University of Science and Technology,Huainan 232001,China)

机构地区：[1]安徽理工大学数学与大数据学院,安徽淮南232001

出　　处：《哈尔滨商业大学学报（自然科学版）》2023年第6期676-684,共9页Journal of Harbin University of Commerce：Natural Sciences Edition

基　　金：深部煤矿采动响应与灾害防控国家重点实验室基金资助项目(SKLMRDPC22KF03).

摘　　要：针对机动车在行驶时,驾驶人、乘坐人员需要佩戴安全带,提出了一种33层卷积神经网络模型进行汽车安全带佩戴识别.介绍了卷积神经网络的卷积算法、池化算法和网络层之间的连接方式并设计了网络结构.针对已有优化算法的准确率与稳定性不足的问题,提出了融入经典动量思想的AWM优化算法.通过AWM优化算法基于车内人员佩戴了安全带和未佩戴安全带的两类数据集对网络的参数进行优化和训练后得到RIVNet模型.实验结果表明,RIVNet模型能够提高汽车安全带佩戴情况检测的精确度,可以高效地进行数据处理和图像的特征提取.以此卷积神经网络模型为基础,基于目标检测算法Faster R-CNN开发出了一款汽车安全带佩戴识别系统.Aiming at the requirement that drivers and passengers need to wear seat belts when a motor vehicle is running,this paper proposed a 33-layer convolutional neural network model for seat belt-wearing recognition.The convolutional algorithm,pooling algorithm,and network layer connection mode of a convolutional neural network were introduced,and the network structure was designed.Aiming at the problem of insufficient accuracy and stability of the existing optimization algorithms,an AWM optimization algorithm integrating the classical momentum idea was proposed.AWM optimization algorithm was used to optimize and train the parameters of the network based on two types of data sets of passengers wearing seat belts and not wearing seat belts,and the RIVNet model was obtained.The experimental results showed that the RIVNet model could improve the accuracy of seat belt-wearing detection,and could efficiently process data and extract image features.Based on this model,a vehicle seat belt-wearing recognition system was developed based on the target detection algorithm Faster R-CNN.

关 键 词：卷积神经网络 网络结构 优化算法 Faster R-CNN 汽车安全带佩戴识别系统 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]