基于多特征融合及PSVM的车辆检测研究  

Study on Detecting Method of Preceding Vehicle Based on Multi-Feature Fusion and PSVM

作　　者：胡均平[1] 邓鹏 李勇成 

机构地区：[1]中南大学机电工程学院,湖南长沙410083

出　　处：《计算机仿真》2017年第12期326-330,共5页Computer Simulation

摘　　要：针对传统单目视觉车辆检测系统准确度及实时性方面研究的不足,提出一种融合多特征的单目视觉车辆检测方法。在车辆假设阶段,结合车辆底部阴影、水平边缘、竖直边缘等多特征融合来产生车辆感兴趣区域,从而产生车辆假设目标;在车辆确认阶段,通过垂直梯度直方图来提取车辆竖直边缘特征,运用于最邻近支持向量机训练器的训练。最后,通过训练好的最邻近支持向量机训练器来对车辆假设进行进一步的确认,从而排除虚假目标。实验结果表明,上述车辆在车辆检测时准确度很高且实时性较好。This paper proposes a monocular vision vehicle detection method with multi -feature fusion. This method aims at the shortcomings of low accuracy and poor real - time effect of the traditional monocular vision vehicle detection system. At the stage of vehicle hypothesis, the vehicle＇s interest area was generated by combining multi - feature vision, such as the shadow of vehicle bottom, horizontal edge and vertical edge, so as to generate vehicle hypothetical target. At the vehicle confirmation stage, the vertical gradient of the vehicle was extracted by the vertical histogram, which was applied to the training of proximal support vector machine trainer. Finally, the trained proximal support vector machine trainer was employed to further confirmation of the vehicle hypothesis, so as to eliminate the false targets. The experimental results show that the vehicle in this paper is very accurate and has excellent real time effect during the detection test.

关 键 词：二维大津法 边缘检测 垂直梯度直方图 最邻近支持向量机 

分 类 号：TP202.7[自动化与计算机技术—检测技术与自动化装置]