不同光照和姿态下的航拍车辆检测方法-北京工业大学学报.PDFVIP

第42卷 第7期 北 京 工 业 大 学 学 报 Vol.42 No.7 2016年 7月 JOURNAL OF BEIJING UNIVERSITY OFTECHNOLOGY Jul. 2016 不同光照和姿态下的航拍车辆检测方法 毛摇 征,刘松松,张摇 辉,孟摇 灿,罗子安 (北京工业大学电子信息与控制学院,北京摇 100124) 摘摇 要:为了解决在不同光照和姿态下的航拍车辆检测准确度低的问题,基于Fourier鄄HOG算法提出了一种航拍 车辆检测方法. 该方法是基于滑动窗口的检测方法. 首先,在处理过程中引入图像预处理,可以将背景区域进行有 选择的剔除,大大节省检测时间和降低虚警率;其次,提取航拍图像基于局部敏感直方图的光照不变性特征;然后, 再提取旋转不变的Fourier鄄HOG特征. 将此特征在线性支持向量机中对车辆目标与非车辆目标进行分类. 在后续 处理阶段,引入非极大值抑制来降低误检目标. 实验结果表明:所提出的车辆检测方法在谷歌地图数据集上进行测 试,其检测准确度较高,且时间消耗低于原始的Fourier鄄HOG检测方法,该方法是一种较为有效的航拍车辆检测 方法. 关键词:车辆检测;局部敏感直方图;Fourier鄄HOG特征;支持向量机;非极大值抑制 中图分类号:TP391郾9 文献标志码:A 文章编号:0254-0037(2016)07-0982-07 doi:10.11936/ bjutxb2015090038 Vehicle Detection From Aerial Photographing Under Different Illumination and Pose MAOZheng,LIU Songsong,ZHANG Hui,MENG Can,LUOZi蒺an (College of Electronic and Control Engineering,Beijing University of Technology,Beijing 100124,China) Abstract:To solve the problem of low detection accuracy of vehicle detection from aerial photographing under different lighting conditions and different postures, a new method based on the Fourier鄄HOG algorithm was proposed. This method was based on a sliding鄄window detection approach. First,image preprocessing,which selectively removed the background region, greatly improved the efficiency of detection and reduced the false alarm rate. Second,illumination invariant features were extracted based on local sensitivehistogramandthentherotationinvariant Fourier鄄HOGfeatureswereextracted. Finally, from the above features, the vehicle and non鄄vehicle were discriminated in a linear support vector machine (SVM) classifier. For post鄄processing,nonmaximum supp