摘要
为了能够对高速公路视频中的违法车辆特征进行提取,需要完成车辆对象的目标检测和特征区域定位工作。在特征区域定位研究中,车标定位是很重要的工作,文中研究了一种直接在车脸图像DCT域中进行车标定位的方法。该方法首先使用Adaboost算法对高速公路视频中的车辆对象进行车脸区域的定位,得到感兴趣的车脸区域图像,然后利用车脸图像中车标纹理和散热器隔栏纹理所具有的不同方向性特点,直接在DCT域中提取车标的横向、竖向、斜向纹理的方向信息,通过区别于散热器隔栏只有横、竖向纹理的特性,采用阈值方法把车标区域从车脸图像背景中分割出来,从而定位到车标区域。实验采用大量不同车标的车辆图片进行处理,结果表明,该算法处理速度较快,且可以有效实现车脸区域的车标定位。
In order to extract illegal vehicles＇ feature in highway video,the target detection of vehicle objects and the localization of characteristics areas need to be completed. In the work of characteristics areas localization,vehicle logo localization is a very important work.A newapproach is proposed for vehicle logo localization directly in DCT domain of face image. Firstly adaptive boosting（ Adaboost） algorithm is used to locate the vehicle face region of vehicle objects and the vehicle face images could be gained. Secondly the vehicle logo ＇s horizontal,vertical,and diagonal texture information in the DCT domain are extracted by the different directional characteristics between vehicle logo ＇s and radiator grille＇s texture in vehicle face images,then threshold method is used to divide the vehicle logo region out of the background image which could rule out the radiator grille＇s horizontal,vertical texture information. A large number of vehicle images with different logos are used for experiments,and the results showthat the processing speed is faster,and the algorithm can effectively achieve the vehicle logo localization of vehicle face region.
出处
《计算机技术与发展》
2016年第4期70-73,共4页
Computer Technology and Development
基金
国家自然科学基金资助项目(61363043)
关键词
车脸
DCT域
车标定位
纹理
ADABOOST
vehicle face
DCT domain
vehicle logo localization
textural features
Adaboost
