Detection Of Traffic Sign Changes Based On Street View Images

With the development of autonomous driving technology,the role of highprecision navigation maps becomes more and more obvious.However,there are challenges in updating high-precision navigation maps.Taking the update of road traffic signposts as an example,the overall mileage of roads in China is large,and the number of traffic signposts is huge.The traditional method of updating the map as a whole will generate excessive costs and a lot of redundant information,which cannot meet the high efficiency of high-precision navigation maps.3.Requirements for real-time updates.There is a need for a method of locally detecting and updating road traffic signs.This article mainly uses street scene images to detect changes in street traffic sign information.In view of the fact that there are many irrelevant changes in street view images,street view image pairs often cannot be directly matched.In this paper,we first extract traffic signs from street view images,and then select appropriate preprocessing methods to process the changed image pairs.In view of the characteristics of the large amount of data,fast update speed and high accuracy requirements of traffic guide signs in high-precision navigation maps,this paper designs a machine learning change detection method that can adaptively discriminate thresholds.Compared with traditional methods,it improves change detection.The accuracy also reduces the time consumption.In order to increase the generalization and automation of the detection model,this paper designs another effective change detection method.The two algorithms have strong anti-interference ability against factors such as light,weather and poor image quality.Specifically,this article studies and develops the following aspects:1.In view of the street scene images collected at different times,the environment changes are complex,and automatic matching cannot be achieved.In this paper,the traffic signs are selected using specific color intervals,and then the Gaussian blur is processed to extract the traffic signs.Among them,for the problem of distortion and different scales of the image of the signpost caused by different parallax and different collectors,the image is rotated and interpolated,and scaled to the same size for storage.2.In order to eliminate the effects of illumination differences and acquisition noise,this paper tests the results obtained by the three filters under different parameters,compromises the signal-to-noise ratio and the degree of distortion,and selects the appropriate filter and its parameters to process the image while removing noise Image features are well preserved.Use the histogram to analyze the pixel value distribution of the image pair,compare the color distribution range of the image,analyze the contrast of the two phases of the image and the degree of the influence of the light,provide data basis for the change detection parameter adjustment,and improve the accuracy of the change detection.3.This paper designs a detection method of traffic signpost change based on image grayscale adaptive SVM.By constructing gray-scaled signpost difference image pairs,input into SVM model for change detection.Experiments show that this method greatly reduces the interference of lighting conditions on the change threshold,and at the same time reduces the complexity of the data and reduces the cost of calculation time.Morphological operators process images,effectively suppressing the interference of noise on the detection results.The detection performance of this method is greatly improved compared with the traditional change detection method.4.This paper designs a change detection algorithm based on capsule network.This method first clusters through SOM,and then uses the capsule network to consider the characteristics of angle,color,and pixel order information to obtain the change detection results.Experiments show that this method exhibits strong generalization performance,which improves the accuracy and automation of change detection in two-dimensional scenes.