Research On Road Traffic Signs Recognition And Lane Detection Under Hazy Weather

With the rapid development of economy and technology,the increase of vehicle ownership brings a series of problems,such as traffic congestion,environmental pollution and so on,which has a negative impact on the development of society.Intelligent transportation system emerges as the times require and develops rapidly,among which lane line detection and traffic sign recognition system,as one of its important components,provides more road information for the driver or the automatic driving control system,and relieves the pressure of the driver to monitor the external environment,which is the key technology to ensure driving safety in automatic driving and auxiliary driving.At present,assisted driving is mainly used in structured road and the technology is not mature,so it is one of the important topics to accurately detect the location of lane line.In addition,the road traffic environment is complex and changeable,and the lighting,weather,occlusion and shooting angle in the road scene put forward high requirements for traffic sign recognition.With the frequent occurrence of haze in northern China in recent years,the image information collected by sensors is fuzzy and degraded.The loss of detail information brings great challenges to traffic sign recognition and lane line detection.In this paper,traffic sign identification and lane line detection in haze weather are studied to solve the picture quality reduction problem caused by haze,and on this basis,lane line rapid detection and scene traffic sign identification are realized.The main work of this paper is divided into the following four parts:Firstly,in the aspect of image defogging,analysis in this paper is based on the defogging algorithm of the prior of dark primary color and Multi-Scale Retinex(MSR).Through the fast guide filter,we refine the estimated transmittance and optimize the estimated value of atmospheric light value,so as to reduce the halo phenomenon caused by the abrupt change of image edge and distance.On this basis,MSR can enhance the image twice in HSV model,and solve the problem of image darkening,and verify the effectiveness of the improved algorithm through experimental comparison.Secondly,in the aspect of lane line detection,the model-based algorithm is comprehensively considered to realize the detection.At first,the lane line model is established,and the road image is preprocessed by grayscale,noise reduction and Canny edge feature extraction.The mask technology is applied to select the area of interest accurately,and then the lane line is detected by Hough transform and least square method.Thirdly,in this paper,based on the analysis of traditional traffic sign recognition algorithm and its application under haze weather,the segmentation algorithm of traffic signs based on color and shape features are respectively studied,and traffic sign recognition algorithms based on Histogram of Oriented Gradient(HOG)and Support Vector Machine(SVM)are applied in haze scenes designs and relatively high recognition rate is achieved on the collected data set.However,there are some false detection and omission phenomena,which have some limitations.At last,analysis in the this paper is object detection algorithm based on deep learning,and the Faster R-CNN network model with high accuracy is applied to the road sign scene in haze weather.Aiming at the phenomenon of small target missing detection,the internal VGG16 network is improved by feature fusion,and experiments are carried out on the collected data set to verify the feasibility of the model,which has a high recognition rate for small targets,and the model has a high robustness,and can accurately locate and identify traffic signs under the harsh conditions of uneven light,complex background,deformation,shadow,fuzzy and so on.