Vehicle Type Classification based on Improved HOG_SVM

Abstract

There are few differences in the characteristics of vehicles and many interference factors in vehicle identification, especially in complex backgrounds. In order to improve the accuracy of image feature extraction and recognition in complex background, a vehicle-types recognition technology based on improved HOG_SVM is proposed in this paper. In order to obtain abundant vehicle identification information, we perform targeted image preprocessing methods such as grayscale stretching and Gaussian filtering on the original image to reduce background interference factors. The HOG feature is then introduced to obtain rich features of the image, and the SVM classifier in machine learning is trained at the output layer by multitasking learning of a large amount of tagged data. Different from the traditional method, the PCA dimension reduction process is used to speed up the recognition of the improved HOG feature, and the method of SVM is used to avoid the classifier from falling into the local minimum. In this paper, the public vehicle dataset is used as the classifier training dataset and test dataset, and the proposed method is verified by experiments.