Vehicle types classification using deep neural network techniques

Traffic flow is one of the most important information elements in intelligent traffic transportation engineering. This study developed a vehicle type classification system using a neural network technique. The architecture of this study is divided into two parts, vehicle pictures are collected first, and then divided into motorcycles, sedans, recreation vehicles, buses and trucks to build a contrast database. The image processing techniques included median filtering and edge detection used to de-noise to improve recognition efficiency. The second stage is processing the previous data stage into the system identification database. All data created by the database were then input into the classifier for calculation. The classification recognition rate was finally obtained. This study uses the Generalized Regression Shallow Learning Neural Network (GRNN), Deep Neural Network (DNN) and Convolutional Neural Network (CNN) classification algorithms. The results show that vehicle classification using the convolutional neural network is better than that obtained with the deep neural network using the restricted Boltzmann machine. Both types of neural networks produced much higher classification than the generalized neural network. The deep learning technique was shown better than the shallow learning approach in this study.