摘要
为有效识别工程车车型,提出一种改进的HOG特征的自动识别算法。使用RPN(region proposal network)自动划分车辆候选区,对每个候选区进行自然度保留的图像增强处理和颜色不变性处理,分别提取NPE-HOG(naturalness preserved enhancement-histogram of oriented gradients)特征和CIV-HOG(color invariant-histogram of oriented gradients)特征,将两者融合得到NC-HOG特征,结合一对一支持向量机实现对压路机、挖掘机、装载机3类工程车辆的车型自动识别。实验对国家电网施工现场的工程车数据库进行测试,对比从原图、Gamma校正、自然度保留图像增强和颜色不变性处理图像上分别提取HOG、LBP(local binary pattern)、SIFT(scale-invariant feature transform)特征对车型识别效果的影响。实验结果表明,改进NC-HOG算法能有效提高HOG特征的识别正确率,3种车型平均识别正确率为93.0%。
To effectively identify engineering vehicles,an improved automatic identification algorithm of HOG features was proposed.Based on RPN(region proposal network)method,candidate regions of vehicle were automatically divided,and each candidate region was processed with natural detail preserving image enhancement and color ineffectiveness processing.NPE-HOG(naturalness preserved enhancement-histogram of oriented gradients)features and CIV-HOG(color invariant-histogram of oriented gradients)features were extracted respectively.NC-HOG was the fusion of NPE-HOG and CIV-HOG.Combined with 1vs1 SVMs,the automatic recognition of three types of engineering vehicles(road roller,excavator and loader)was realized.Experiment tested the engineering vehicle database of the State Grid construction site.Effects of the characteristics of HOG,LBP(local binary pattern)and SIFT(scale-invariant feature transform)which were extracted from the original image,gamma correction,natural degree preserving image enhancement and color invariance processing image on vehicle recognition were compared.Experimental results show that the NC-HOG algorithm can effectively improve the recognition accuracy of HOG features.The average recognition accuracy of the three models is 93.0%.
作者
罗亮
吕俊杰
李涛
张劲
刘俊勇
刘友波
LUO Liang;LYU Jun-jie;LI Tao;ZHANG Jin;LIU Jun-yong;LIU You-bo(Construction Department,Aba Power Supply Company,State Grid Sichuan Electric Power Company,Aba 623200,China;Tianfu New Area Power Supply Company,State Grid Sichuan Electric Power Company,Chengdu 610041,China;College of Electrical Engineering,Sichuan University,Chengdu 610065,China)
出处
《计算机工程与设计》
北大核心
2021年第11期3164-3173,共10页
Computer Engineering and Design
基金
国家自然科学基金项目(51977133)
国家电网四川省电力公司科技基金项目(52191918003L)。