Abstract
Vehicle recognition is a vital part of the advanced driver assistance system (ADAS). In this paper, we propose a multi-features fusion (MFF) method combined with convolutional neural network (CNN) for vehicle recognition, which can enhance the vehicle recognition relative to the single feature and improve the accuracy of vehicle recognition. Our method firstly trains the CNN using different convolution kernels, network layers and feature maps to obtain the optimal CNN model, and extracts all feature vectors of the hidden layer in an image, namely CNN features. At the same time, the histograms of oriented gradients (HOG) features and the principal component analysis (PCA) features of the vehicle are also extracted from the image separately. Then we fuse these features with different normalization methods to obtain a more distinguishable comprehensive feature. Finally, we can recognize vehicle in the image by support vector machine (SVM). Our experimental result shows that the proposed MFF method is superior to CNN+HOG+SVM, HOG+PCA+SVM and PCA+CNN+SVM feature fusion methods, and the accuracy is further improved from 96.63% to 98.00%.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call