基于多特征提取和SVM参数优化的车型识别

Vehicle Recognition Based on Multi-feature Extraction and SVM Parameter Optimization

提出了一种基于多特征提取和支持向量机(support vector machines,SVM)参数优化的车型识别方法,此方法解决了采用单一特征容易受到光照、天气、阴影等环境影响的问题,并且可以对运动中的车辆进行车型识别。首先,采集车辆样本并进行图像预处理,提取车辆的几何特征、纹理特征和方向梯度直方图(histogram of oriented gradient,HOG)特征;其次,将提取的多种特征量进行组合测试,并与单个特征量的测试结果进行比较;最后,采用粒子群算法优化SVM的参数并使用优化的SVM参数进行运动车辆的车型识别。实验结果表明:提出的多特征提取和SVM参数优化相结合的车型识别方法能够取得很好的识别效果,识别率达到90%以上。

A kind of vehicle recognition method which was based on multi-feature extraction and support vector machines （SVM） parameter optimization is proposeal. Many kinds of problems that used the single-feature can be influenced by those factors such as light, weather and shadow, etc. Those problems could be solved by our method. In addition, our method can identify the moving vehicle model. At first, the samples of vehicle are collected and begin the process of image preprocessing, a variety of features will be extracted, including geometric features, texture features and histogram of gradient features. The second, combining and testing the various features, then the results with the results of single-feature testing are compared. At last, preparing for the recognition of the vehicle by SVM which was optimized by Particle Swarm Optimization（PSO）. The experimental results show that the method which is put forward can achieve a good recognition results. The recognition rate can reach more than 90%.