Road Traffic Sign Detection Under Natural Scene Based On Deep Learning

With the continuous increase of car ownership in China,especially the arrival of the era of private car consumption and the development of driverless technology,driving safety has become a hot topic in the development of people's livelihood.Therefore,many scholars have come up with the concept of “intelligent transportation”,aiming at perceiving road information,assisting driving,avoiding risks and ensuring safe driving.Topics such as the effective detection of traffic signs and the result feedback to drivers have become hot in the field of intelligent transportation and driverless technology.Previous traffic sign detection is mostly based on its color,shape,texture,etc.but due to the variety of traffic signs and the relative complexity of natural environment,previous detection models have low ability to process images,hence its impracticality.In recent years,with the development of deep learning technology,the deep neural network model,which has good detection effect and high accuracy,has been applied to traffic sign detection.Therefore,this paper,based on deep learning theory,mainly studies the traffic sign detection in natural scenes.Here lists our main research work:(1)The data set needed in this study is established according to guidelines for the establishment of traffic signs in China.It is then further preprocessed for follow-up experiments.(2)It contrasts the appearance features of different kinds of traffic signs,following which traffic sign detection algorithms based on color space and shape features are studied respectively.What's more,a HOG feature detection algorithm based on color and shape features is designed via combining the advantages of the two detection algorithms mentioned above.Although its detection accuracy on the collected data set is up to 0.534,time consumed is relatively long,hence its poor robustness and practicability.(3)It analyzes the target detection algorithm based on deep learning theory,which is further applied to the task of traffic sign detection in natural environment.Faster R-CNN network is finally chosen as the detection network in this paper after comparing both the advantages and disadvantages of deep learning frameworks of R-CNN series.The feasibility of this model is verified by changing the anchor ratio.Based on the improved Faster R-CNN network,a traffic sign detection algorithm in natural scenes based on feature fusion is proposed.The accuracy of the detection is 0.904 on the collected data set,indicating its practicality.It is also fast and robust.(4)Considering its function,a traffic sign detection system based on in-depth learning is designed to detect traffic signs in various natural scenes.