基于改进KNN-SVM的车辆图像光照检测模型

发布时间：2018-03-14 04:25

本文选题：车辆交通图像　切入点：光照特征　出处：《计算机工程与应用》2017年24期 　论文类型：期刊论文

【摘要】：为了准确检测出车辆交通图像的光照类型,从而有针对性地矫正不同光照以减少其对车牌定位的影响,提出了一种基于改进K近邻和支持向量相融合(KNN-SVM)的车辆图像光照检测方法。首先融合了HSV空间亮度特征、灰度直方图特征和投影直方图特征作为车辆图像的光照特征,然后改进传统KNN-SVM中距离计算方法,定义为每类待检测样本到属于该类支持向量的距离,并在采集的全天候不同光照车辆图像上进行检测验证。实验表明,改进KNNSVM将阈值获取时间提前,避免了传统KNN-SVM对超平面附近样本先SVM检测再KNN检测的重复检测,不仅降低了算法复杂度和运行时间,且检测准确率高于传统KNN-SVM和单独使用KNN或SVM时的值,最高达到了99.67%。
[Abstract]:In order to detect the illumination type of vehicle traffic image accurately, and correct different illumination in order to reduce its influence on license plate location, A vehicle image illumination detection method based on improved K-nearest neighbor and support vector fusion (KNN-SVM) is proposed. Firstly, the luminance feature of HSV space, gray histogram feature and projection histogram feature are combined as illumination features of vehicle image. Then, the distance calculation method in traditional KNN-SVM is improved, which is defined as the distance between each kind of samples to be detected and the support vector of this class, and the detection verification is carried out on the collected all-weather and different illumination vehicle images. The experimental results show that, The improved KNNSVM improves the threshold acquisition time in advance, avoids the repeated detection of the traditional KNN-SVM to samples near the hyperplane by first SVM detection and KNN detection, which not only reduces the algorithm complexity and running time, but also reduces the algorithm complexity and running time. The accuracy of detection is higher than that of traditional KNN-SVM and using KNN or SVM alone, and the highest is 99.67.
【作者单位】：北京信息科技大学计算机学院计算机系统开放实验室;
【基金】：宁波市镇海区2016年引进高层次人才创业项目
【分类号】：TP18;TP391.41

【相似文献】