摘要
行人特征提取是行人识别中关键步骤之一。传统的做法是分别从彩色图像的每个颜色通道(即R、G、B通道)中提取特征描述子(如方向梯度直方图(HOG)、局部二值模式(LBP)等),最后合并成一个特征向量。传统方法不足之处在于难以兼顾不同颜色通道之间的关联信息。为此,采用四元局部排序二值模式(QLRBP)运算方法从彩色图像中提取局部特征描述子。与传统方法不同的是,该方法是一种整体的方法,是在彩色图像的四元数表示空间而非3个颜色通道中分别提取LBP特征。首先,将从车载摄像头中采集的彩色图像通过四元数转换获得其四元数表示;然后,对四元数空间中图像进行CTQ变换,并计算变换后的图像相位;最后,在每个四元数相位上进行LBP运算,并生成行人彩色图像的局部特征描述子。QLRBP能够同时处理所有的颜色通道,因此能够同时包含三通道之间的关系。在行人判定方法上,本文首先提取所有正负样本的QLRBP特征,并使用K-最近邻(K-NN)算法训练分类器。在INRIA数据库上测试表明,QLRBP描述子对于彩色行人的检测比其他常用的特征描述子(如HOG特征,LBP特征)具有更高的精确度,性能接近当前先进的行人识别方法。
Pedestrian feature extraction is one of the key steps in pedestrian recognition.The traditional method of pedestrian recognition is to extract feature descriptors(such as HOG,LBP)from each color channel(R,G,B channels),Finally merge into a feature vector.it is difficult to take account of the correlation information between different color channels.In this paper,we use a holistic approach to extract local feature descriptors from color images,which is called quaternionic local ranking binary pattern local descriptor(QLRBP).Unlike traditional methods,this method extracts LBP features from the quaternionic representation space instead of the three color channels.First,Encoding a color pixel using aquaternion to get the quaternionic representation(QR)of the color image which collected from a vehicle mounted camera.Then,Applying a Clifford translation to QR of the color image.Finally,Performing a local binary codingon the phase of the transformed result to generate local descriptors of the color image.QLRBP is able to handle all color channels directly in the quaternionic domain and include their relations simultaneously.In the method of pedestrian recognition,the positive and negative samples are collected first.The QLRBP features are extracted from all the samples,and the K-nearest neighbor algorithm is used to train the classifier.The method is tested on the INRIA pedestrian database and shows that it is better than other features,such as HOG features and traditional LBP features.Performance approach to the current advanced method of pedestrian recognition.
作者
田甜
黄影平
胡兴
慈文彦
Tian Tian;Huang Yingping;Hu Xing;Ci Wenyan(School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China)
出处
《电子测量技术》
2019年第4期117-122,共6页
Electronic Measurement Technology
关键词
行人特征提取
四元数
四元数局部排序二值模式
行人识别
K-最近邻算法
pedestrian feature extraction
quaternion
quaternionic local ranking binary pattern
pedestrian recognition
K-nearest neighbour algorithm