Research On Deep Learning Based Vehicle Type Recognition System

Intelligent transportation system is an important part and also the future trend of modern transportation.Automatic vehicle type recognition is an important research direction in the intelligent transportation system.In many scenarios such as vehicle parking management,road traffic monitoring and traffic statistics,vehicle type recognition plays an important role and has a wide range of applications.Typically vehicle type recognition systems make real-time analysis and processing of video surveillance data of traffic videos using image processing and pattern recognition techniques.In recent years,there has been a qualitative breakthrough for image processing and recognition technologies.Deep Learning technology including CNN(Convolutional Neural Network)and DBN(Deep Belief Network)have a strong self-learning ability,and it has made breakthrough achievements in image recognition.At the same time,Deep Learning is also applied in speech recognition and natural language processing areas.CNN has excellent performance in image feature extraction and recognition,then it is widely used in various fields of image recognition.Firstly,we make process to traffic videos with technologies which combines the traditional video processing and morphological image processing technologies to extract the moving vehicles,and do standardization of the vehicle images.Then construct the original training data set with the images extracted from the video and train the deep convolutional neural network designed for vehicle type recognition.The trained neural network works as a background module of vehicle type recognition system,the constantly incoming foreground vehicle images are recognized by it which gives the results of vehicle recognition and the probability of the result.At last combine the vehicle recognition results with spatial and temporal information in the video to do vehicle tracking.This study is based on the windows platform and the program is implemented in python language,the results show that it performs well in real-time multi-car video vehicle type recognition and tracking combining deep learning with traditional video and image processing technologies.