基于Zernike不变矩和HOG梯度直方图复合特征的车标识别
首发时间:2014-12-11
摘要:车标识别技术是智能交通系统中的一个重要的研究领域。对交通监控录像中车牌污损、遮挡的肇事车辆信息的辅助确认是现阶段智能交通系统中的一个重要问题,车标作为一个关键特征,可以起到较好的辅助判别的作用,具有重要的实用价值。车标识别是指通过计算机视觉、图像处理与模式识别等方法从车辆图像中提取车标信息从而获得机动车辆品牌信息的一种实用技术。车标识别系统中主要包括两个关键部分:车标定位和车标识别。本文研究的是将已定位好的车标进行有效识别,并针对车标识别方法,在前人研究的基础上对Zernike不变矩和HOG梯度直方图特征进行结合形成复合特征在SVM分类器下用于车标识别,提出了一种复合式特征的车标识别方法。实验结果表明该复合特征比单一的Zernike不变矩和HOG梯度直方图下的特征在抗干扰性如抗旋转,光照,噪声方面具有更强的适应性和鲁棒性,能够有效降低车标识别的误检率。
关键词: 车标识别 Zernike不变矩 HOG梯度直方图 支持向量机
For information in English, please click here
Vehicle logo recognition based on the hybrid feature of Zernike moments and HOG gradient histogram
Abstract:Vehicle logo recognition technology is an important area of the intelligent transportation system. When license plate is defaced or veiled artificially in traffic surveillance video, it's hard to confirm the information of the car, this is an important problem in the intelligent transportation system. The vehicle logo, as a key feature of a car, can play a better role in the secondary discrimination.Vehicle logo recognition aims to recognize the logo information of the vehicle by coupling the methods of computer vision, image processing and pattern recognition. Vehicle logo recognition system consists of two key components: vehicle logo detection and vehicle logo recognition .This paper presents a method that can effectively identify the vehicle logo which has been located. In this paper, a new vehicle identification feature is proposed by combining Zernike moments and HOG gradient histogram, which can used as the input of SVM classifier to classify vehicle logo. Experimental results show that the hybrid feature has stronger adaptability and robustness than Zernike moments or HOG parallel gradient histogram features in terms of resisting interference such as resisting rotation, light, noise which can effectively reduce false detection rate of vehicle logo recognition.
Keywords: Vehicle Logo Recognition; Zernike Moments;HOG Gradient Histogram Features;SVM
论文图表:
引用
No.4621824998057141****
同行评议
勘误表
基于Zernike不变矩和HOG梯度直方图复合特征的车标识别
评论
全部评论0/1000