CNN与人工特征提取快速识别斑马线的方法  

Research on the method of CNN and artificial feature extraction for rapid identification of zebra crossing

作　　者：熊平[1] 胡彩霞 周欣星 

机构地区：[1]中南大学生物医学工程系,湖南长沙410083 [2]湖南中医药大学第一附属医院,湖南长沙410000

出　　处：《电子设计工程》2018年第3期189-193,共5页Electronic Design Engineering

摘　　要：斑马线识别方法的研究对车载和导盲系统具有极其重要的作用。为了解决目前识别斑马线的方法精确度低、所需时间长等问题,本文提出了一种基于卷积神经网络与人工特征提取相结合的快速识别斑马线方法,卷积神经网络能自动检测出所有的符合斑马线特征的区域,再利用梯度极大值比较等人工特征提取方法进一步精确检测范围。通过实验分析与数据训练,结果表明这种方法运算速度能在60 ms内,且识别出的斑马线精确度高达96%。Study on the zebra crossing recognition plays an important role in vehicle and guide system.In order to solve the problems of low precision and long time needed to identify the zebra crossing,this paper presents a rapid method of detecting zebra crossing,through combining the method of convolutional neural network and the method of artificial feature extraction.Convolutional neural network can automatically detect all possible region,and using the gradient maximum comparison of artificial feature extraction method can further accurate detection range.Through the experimental analysis and data training,the results show that this method can be used in 60 ms,and the accuracy of the zebra crossing is up to 96%.

关 键 词：卷积神经网络 人工特征提取 梯度极大值 斑马线 检测识别 

分 类 号：TN911.73[电子电信—通信与信息系统]