Vehicle logo recognition based on deep learning architecture in video surveillance for intelligent traffic system

Vehicle logo acquisition or recognition has been a popular study field in intelligent traffic system for the latest decade. In this paper, a vehicle logo recognition method based on CNN (Convolutional Neural Network) is introduced. In the experiment, two classification methods with different feature extraction methods were applied. However, these two classification methods were utilized the same SVM classifier in classification procedure. The experiment uses input images with height of 140 pixels and width of 100 pixels. The samples include 26 classes, which are derived from 16 different vehicle brands. The comparative experiment results indicate the CNN performs better in recognizing vehicle logos; the average accuracy rate of 26 classes achieves 99.23%, which is 8.61% higher than the other approach based on SIFT, whose accuracy rate is 90.62%.