Research On Traffic Sign Detection And Recognition Based On Machine Learning

Traffic sign recognition system TSR(Traffic Sign Recognition)is one of the important subsystems of ADAS.The TSR system provides drivers with warning and guidance information by identifying traffic signs,and intelligently guarantees the orderly and standardized driving of vehicles.Aiming at the key problems of multi-scale traffic signs in the natural environment,which are prone to missed detection,detection delay,and low recognition accuracy,this paper focuses on machine learning-based traffic sign detection and recognition methods.The main research contents of this article are as follows:(1)A method of constructing traffic sign data set and image preprocessing is proposed.Aiming at the common open source datasets of traffic signs that contain few types of traffic signs and large differences at home and abroad,we have constructed three types of domestic traffic sign datasets including instruction signs,warning signs,and prohibition signs.There is a large difference in the number of sample label boxes.Image preprocessing techniques such as multi-channel fusion are used to equalize the data set,thereby providing sample support for subsequent training and testing of the traffic sign detection and recognition algorithm in this paper.(2)The research proposes a traffic sign detection method based on feature enhancement and improved SSD.Aiming at the problem of low accuracy of small-scale traffic sign detection in the conventional SSD target detection framework,after analyzing and comparing the impact of different basic networks on SSD performance,a VGG-16 network with higher comprehensive performance was selected as the basic network of the SSD;Considering the imbalance between the detail information and semantic information of small-scale traffic signs on the high and low feature layers of SSD networks,this paper first uses a multi-level feature fusion method,and then combines the SE module to assign channel feature weights,thereby improving the small-scale traffic sign Detection accuracy;the improved non-maximum suppression algorithm can not only simplify the detection frame,but also avoid greedy suppression.The experimental results show that the improved SSD-based traffic sign detection method based on feature enhancement effectively reduces the rate of missed detection and false detection of traffic signs,and significantly improves the accuracy of traffic sign detection.(3)The research proposes a traffic sign recognition method based on deep visual feature learning.The recognition principle of the convolutional neural network is studied.For the type recognition network based on the conventional convolutional neural network model,only the category information is learned.This chapter adds a visual characteristic learning network to the conventional convolutional neural network to change the shape of traffic signs The three visual characteristics elements,color,graphic content,and category information are used as network learning content at the same time.After learning the semantic information described by the visual characteristics,the network can more easily understand the traffic signs,and further improve the accuracy of type recognition.(4)Designed and implemented a set of traffic sign detection and recognition software system based on the improved algorithm in this paper.The test results of the software system show that the system has an accuracy rate of more than 95% for multi-scale traffic sign recognition under natural environment,accuracy and real-time performance are expected.