摘要
为探究汽车驾驶人非安全驾驶行为的识别问题,在简要分析现有驾驶人行为识别方法的基础上,提出一种基于卷积神经网络(CNN)的驾驶人行为识别方法,分析CNN前向传播与反向传播过程,给出处理驾驶人行为识别问题的CNN网络架构。结果表明:用该方法可识别,其平均识别率达97.13%,相对于传统提取方向梯度直方图特征(HOG),并用随机森林(RF)分类的算法,该方法的识别率平均提高了3.62%。
In order to explore identification of unsafe driving behaviors of car drivers,concrete studies were carried out on CNN-based driver behavior recognition algorithm building on brief analysis of existing driver behavior recognition methods.CNN forward propagation and back propagation processes were explored and a CNN network architecture that deals with driver behavior recognition was presented.The results show that this method achieves an average recognition rate of 97.13%on state-farm driver behavior dataset,and compared with traditional algorithm,it has improved 3.62%on average in extracting histogram of oriented gradient(HOG)feature and using random forest(RF)classification for identification.
作者
徐丹
代勇
纪军红
XU Dan;DAI Yong;JI Junhong(School of Mechatronics Engineering,Harbin Institute of Technology,Harbin Heilongjiang 150001,China)
出处
《中国安全科学学报》
CAS
CSCD
北大核心
2019年第10期12-17,共6页
China Safety Science Journal
基金
机器人技术与系统国家重点实验室自主课题(SKLRS201705A).
关键词
驾驶人行为识别
卷积神经网络(CNN)
前向传播
反向传播
深度学习
驾驶安全
driver behavior recognition
convolutional neural network(CNN)
forward propagation
back propagation
deep learning
driving safety
