Research On Road And Sign Detection Method Based On Deep Learning

In recent years,artificial intelligence has become a major trend,and smart cars equipped with advanced assisted driving technology have become the focal point of the automotive industry,and environment perception is an extremely important part of advanced assisted driving technology,and the detection of traffic signs and lane lines is an integral part of it.In this paper,we design a traffic sign and lane line detection algorithm based on deep learning,and analyze the performance of the algorithm through experimental verification.The main work of this paper is as follows.(1)Improvement of YOLOv3 algorithm.Firstly,the difficulties of traffic sign detection are presented,and then several algorithm models are listed,and the differences of these algorithms are derived through comparative analysis.Then,considering the characteristics of small traffic sign targets,combined with the requirements for real-time detection tasks in real scenarios,so the YOLOv3 network is chosen to be optimized and improved,so that YOLOv3 can detect some small target traffic signs with better results,and finally,the The optimized algorithm is experimented using CCTSDB dataset,and the results show that the detection accuracy has been improved to some extent.(2)This paper conducts a study of lane line detection by using the Tusimple dataset.The accuracy of lane line detection is further enhanced by creating the dataset required for lane line detection,transforming the collected data and the annotation information of the dataset into a format,and finally organizing the structure of the dataset required for this paper.(3)In this paper,Lane Net algorithm is selected as the basis of lane line detection network model based on deep learning.The network mainly consists of an encoding-decoding structure,in which encoding is to extract feature information from the images in the network by convolutional operation and to reduce the dimensionality of the images;decoding is to segment the images and to restore the dimensions of the images.Then the Mean-Shift algorithm is used for clustering,and finally the pixel points are fitted using the least square method to finally achieve the detection of lane lines with high accuracy.