摘要
为研究融合多类特征形成的新特征对车辆识别准确率的影响,首先对图像的方向梯度直方图(HOG)特征、不变矩特征和灰度共生矩阵特征进行特征提取,并对HOG特征采用主成分分析法(PCA)进行降维;利用极差变换法对三类特征进行归一化处理并采用线性融合法进行融合构成新的特征。实验结果表明,基于上述三类单一特征的车辆识别准确率分别为51.39%,59.72%和75%,识别准确率较低;基于融合后形成的新特征的车辆识别准确率达到了97.22%,识别准确率有了较大提高,验证了多类特征融合在提高车辆识别准确率方面的有效性,对研究道路交通安全具有重要意义。
In order to study the influence of the new features formed by the fusion of multi⁃class features on the accuracy rate of vehicle identification,the image features of histogram of oriented gradients(HOG),invariant moment and gray level cooccurrence matrix are extracted and the principal component analysis(PCA)is used to reduce the dimension of the HOG feature,and the three kinds of features are normalized by using the extreme difference transformation method and are fused by using the linear fusion method to form the new features.The experimental results show that the accuracy rates of vehicle identification based on the above three kinds of single features is 51.39%,59.72%and 75%respectively,which means they are all low;the accuracy rate of vehicle identification based on the new features formed by fusion is 97.22%,which indicates it is greatly improved.This research result verifies the effectiveness of the method of multi⁃class features fusion in improving the accuracy rate of vehicle identification,and the method is of great significance to the study of road traffic safety.
作者
王左帅
谭德荣
徐艺
侯汝红
王立志
WANG Zuoshuai;TAN Derong;XU Yi;HOU Ruhong;WANG Lizhi(Shandong University of Technology,Zibo 255000,China)
出处
《现代电子技术》
北大核心
2020年第1期31-34,39,共5页
Modern Electronics Technique
基金
山东省自然科学基金（ZR2016EL19）
山东省自然科学基金（ZR2018PEE016）
中国博士后科学基金（2018M632696）
关键词
车辆识别
特征提取
特征降维
特征融合
归一化处理
交通安全
vehicle recognition
feature extraction
feature dimension reduction
feature fusion
normalized processing
traffic safety
