Font Size: a A A

Research On Road Traffic Sign Detection Method Based On Deep Learning

Posted on:2021-03-13Degree:MasterType:Thesis
Country:ChinaCandidate:Y L ZouFull Text:PDF
GTID:2492306107974369Subject:Engineering (in the field of vehicle engineering)
Abstract/Summary:PDF Full Text Request
Road traffic occupies a leading position in the way of travel in modern society.Road traffic signs can transmit useful traffic information to drivers and occupants in a timely manner,ensure the safety of the vehicle driver’s personal and property,and prevent the occurrence of unexpected road conditions.Therefore,the accurate detection of the traffic signs by the intelligent vehicle environment awareness system is of great significance.It provides technical guarantee for the safe driving of intelligent cars.Traffic sign detection requires accurate and high-performance detection and recognition of traffic sign signals under the conditions of darkness,backlighting,bad weather and complex backgrounds.The current traffic sign detection algorithm has outstanding problems of missed detection and false detection of targets.The algorithm is simple and the detection accuracy is low,which cannot meet the requirements for accurate positioning of traffic signs.The detection algorithm with high accuracy has high complexity and overfitting.The lack of poor real-time performance makes it difficult to apply to vehicle-mounted systems.In view of the above problems,this paper further studies the algorithm for identifying road traffic signs,and proposes a road traffic sign detection method based on deep learning.Firstly,through the analysis of traffic sign detection and recognition,the Chinese traffic sign CCTSDB dataset is selected as the research object for manual annotation and filtering and noise reduction processing,and the traffic sign detection algorithm program based on Visual Studio2015 C ++ platform to build a machine vision model.Secondly,for the problem of poor accuracy of smart car traffic sign detection,a YOLO v3 traffic target detection classifier system based on regression end-to-end connection in keras environment is constructed,and the data set marked in this article is trained using deep network,and The driving camera video collected on real urban vehicles on Chongqing urban roads and inner ring highways is verified.The results show that the detection of visual sensors based on the YOLO v3 convolutional neural network learning algorithm has good accuracy and real-time performance.Finally,deeply study the application of traffic sign detection based on Faster R-CNN algorithm under ZF net and VGG16 net,use the feature transfer connection between the upper and lower layers of the convolutional neural network structure,improve the Faster R-CNN network,and reduce the feeling of the feature map field,and verify the feasibility of the algorithm.Through the comparative analysis of the detection experiments of multiple sets of algorithms,the results show that the improved Faster R-CNN network has better performance for traffic sign detection,which can reduce the problem of missed detection and false detection of the target,and accurately obtain traffic The location and classification information of signs can further improve the accuracy and environmental adaptability of traffic sign detection algorithms,and provide reliable information for assisted driving such as vehicle motion path planning and control decisions.
Keywords/Search Tags:Deep Learning, Convolutional Neural Network, Traffic Sign Recognition, Faster R-CNN Algorithm, YOLO v3 Algorithm
PDF Full Text Request
Related items