Research On Machine Vision And Deep Learning Technology For Intelligent Transportation System

With the continuous improvement of living standards,people’s demand for transportation is increasing.The increase of various types of transportation has not only promoted economic development,but also greatly facilitated people’s travel.However,the subsequent traffic safety and congestion problems have become the focus of social conflicts.Based on the above background,establishing an intelligent transportation system(ITS)is an effective way to solve the above social problems.With the development of artificial intelligence and other technologies,we have provided us with a better solution to data acquisition and data processing in ITS key technologies.At present,with the continuous in-depth research and application of machine vision and deep learning technology in ITS,as the most concise and effective means of guaranteeing and controlling vehicles in ITS,license plate information recognition technology and traffic information detection and recognition technology have been extensively studied.This paper separately researches the license plate information recognition technology and traffic information detection and recognition technology in ITS.First of all,this article is based on the relevant technical theories of the research content,taking the deep learning model framework as the entry point,and deeply analyzes the construction and training of the deep learning model,and then proposes guiding ideas and theoretical reserves for the subsequent improvement of the algorithm model of this article.Secondly,through the analysis of the deep learning framework,in the license plate information recognition system part,relying on the system function requirements,taking the system software design ideas as the main line,and taking the software realization result as the goal,the software modules are laid out and designed to achieve improvements.Research on the algorithm of license plate character recognition.The first step is to preprocess the license plate image.The preprocessed license plate image is located by the license plate location algorithm that combines color information and edge information,and the improved license plate character segmentation algorithm based on vertical projection is used to segment the located license plate characters.Then focus on the analysis of the traditional license plate character recognition algorithm and the license plate character recognition algorithm based on deep learning.Because the deep learning algorithm has more obvious advantages in recognition accuracy and robustness,this paper uses an improved Le Net-5 license plate character recognition algorithm to recognize the segmented license plate characters.Finally,it is analyzed and tested to meet system application requirements.Finally,through the research on the license plate character recognition algorithm based on deep learning,accumulated experience in deep learning model construction and training.In the part of the traffic information detection and recognition system,since the traditional detection and recognition algorithms process the lane lines and traffic signs that they recognize separately,this paper uses deep learning algorithms to detect and recognize lane lines and traffic signs at the same time.Aiming at the feature that the YOLOv3 network model can be used for multi-type target detection and recognition,based on the Tiny-YOLOv3 model,using Tsinghua Tencent’s 100 K China traffic information database,the traffic information is detected by using the improved Tiny-YOLOv3 network algorithm Recognition.The experimental analysis results show that the improved algorithm is effective for the detection and recognition of multiple types of target traffic information including lane lines and traffic signs.