Vehicle Type Recognition Based On Deep Learning

Object recognition is one of the most important research fields of computer vision.Vehicle type recognition, as one of the extensions of object recognition, is indispensable inintelligent traffic systems, considering today’s complex urban traffic conditions. This paperfocuses on figuring out how to use computer vision methods for vehicle type recognition andclassification.We first summarized the popular features and algorithms used in object recognition andclassification. Deep analysis on those useful image descriptors is given in this paper. And theproperties and relationship of those descriptors are also analyzed. We summed up theirextraction algorithms as well. In addition, we gave an overview of the most usefulclassification algorithm used in this field. The theory basis of deep neural network is studied,and different learning methods which are used in deep neural network are compared. We alsoexplained the way to train convolution neural network. Finally, we choose k-means to learnthe image feature and build up the recognition system based on convolution neural networkmethod.We collected a total of7158images covering30types of vehicle to do the experimentsto verify the characteristic of the deep learning network on vehicle recognition. And we alsoused improved SIFT matching method to classify the same image set. During theexperiments, deep learning network achieved94%accuracy, after compared to SIFTmatching results, we can come to a conclusion that the deep learning can be used to dovehicle type recognition.