Performance evaluation of HOG and Gabor features for vision-based vehicle detection

This paper investigates the performance of image features for vehicle classification. We focused on two important image features which have been widely used for vehicle detection. These features are the Histogram of Oriented Gradient (HOG) and the Gabor features. Although there are several literature proposed these features for vehicle classification, it is very hard to make a fair comparison from their published results since they were tested using different data sets and performance matrices. This paper compares the performance of these two features under the same experimental setups. The efficiency of the features in combination with three popular classifiers, namely Support Vector Machines (SVM), Multilayer Perceptron Neural Network (MLP) and Mahalanobis distance classifiers were systematically investigated. The experiment results show that the combination of HOG feature with SVM classifier produced the best result. The processing time required for HOG feature´s extraction and classification is also considerably shorter compared to Gabor feature.