摘要
针对道路视频监控中车型识别的问题,为准确定位车前脸,提出了一种基于车前脸梯度方向直方图的识别算法。用形态学粗定位和投影细定位算法提取视频中车前脸区域,准确定位车前脸,能提高全局特征算法的识别效果。将车前脸图像的梯度方向直方图特征作为识别初始特征,采用线性判别分析算法进行特征提取,降低特征维数,提高识别速度。基于集成学习的思想,对车前脸进行网格分割,各子区域训练得到的分类器生成集成分类器,提高车型识别率。建立了15种车系80种车型的车前脸图像库进行实验,实验结果表明,上述方法的车型正确识别率为93.5%。
Using Histograms of Oriented Gradients feature of a vehicle frontal face, this paper presented a vehicle - type recognition method for vehicle - type recognition in road video surveillance, which improves the recognition rate. To begin with, a vehicle frontal face was determined by combining the method of mathematical morphology and projection method. Then the Histograms of Oriented Gradients features were extracted from the image of the front view of the vehicle and processed by using linear diseriminant analysis algorithm. Finally, based on the idea of ensemble learning, vehicle frontal face was divided into grids, and every sub area was trained to integrate an ensemble classifi- er. Experiments were carried out based on a dataset of vehicle frontal face images including fifteen vehicle makes and eighty models. The experimental results show that the correct recognition rate of the proposed method is 93.5%.
出处
《计算机仿真》
CSCD
北大核心
2015年第12期119-123,共5页
Computer Simulation
基金
南京航空航天大学研究生创新基地(实验室)开放基金(kfjj201428)
国家自然科学基金(61371170)
关键词
梯度方向直方图特征
车型识别
车前脸
线性判别分析
集成学习
Histograms of oriented gradients feature
Vehicle - type recognition
Vehicle frontal face
Linear dis- crimination analysis
Ensemble learning
