Research And Implementation Of Chinese Traffic Panel Recognition And Text Extraction System

In recent years,artificial intelligence technology has been rapidly developed and gradually applied to the field of intelligent transportation.It brings more and more convenience to people's lives.The traffic panel is an important part of intelligent transportation,which contains important information such as road conditions,warnings and safety.The existing traffic panel positioning methods based on color segmentation are only for the monochrome background,while China's traffic panels are mainly with blue and green as background color,so the existing methods are not applicable to China's current situation.This thesis proposes a scheme for Chinese traffic panel location and recognition under natural scenes,and designs and implements a traffic panel location recognition system.The details are as follows:First of all,This thesis proposes a traffic panel positioning scheme based on HSV and watershed algorithm,which uses the model of"filter + HSV color model + geometric features"to achieve the positioning of the traffic panel under natural scenes.The method firstly uses the distribution of image gradation,combined with the watershed image segmentation algorithm,to filter the blue sky region and green plant region in the image that easily interfere with the traffic panel positioning.Then set the optimal threshold for the HSV to preserve all of the traffic panel areas in the image.Finally,using the geometric characteristics of the domestic traffic panel,the candidate areas of all traffic panels are screened out.Through experiments,we prove the effectiveness of the method.On our dataset,we can locate about 77%of Chinese traffic panels,which also proves the effectiveness of the filter.Secondly,In this thesis,a method for extracting the features of the edge regions of traffic panels is proposed.We combine the HOG features with the extracted features and get the traffic panel recognizer based on the ANN training.The survey found that the domestic traffic panel has a white border near the edge,which can more effectively identify the traffic panel area.On our dataset of the traffic panel,through a large number of experiments,our method reduced the average error value by 2 points and the accuracy rate by about 0.4 percentage points.Finally,this thesis designs and implements a Chinese traffic panel recognition and text extraction system.The system consists of two parts:the client and the server.The client system provides the user with the location and recognition function of the traffic panel.At the same time,the user can score the recognition result and upload it to the user.The server helps the server to process the data again and continuously form a relatively reasonable data set.At present,China still lacks the dataset of Chinese traffic panels,so this system is of great value for establishing a relatively complete Chinese traffic panel dataset.The system uses Google's open source Tesseract technology to train Chinese characters involved in the system,and finally realizes the extraction of Chinese text information.