Video And GPS-based Precision Positioning Algorithm For Mileposts

While China is developing and building highways on a large scale,the management problems of the huge road network are becoming increasingly prominent.In the process of road management,the effective positioning of road information is vital to road disease management and accident rescue.Mileage piles,the positioning benchmark of the highway management system,specifically describes the current road mileage status.It is a scientific and effective highway positioning method.In 2018 the Ministry of Transportation began the process of renumbering the national highway network,which also meant that a large number of mileposts were renumbered,installed,and needed to be updated with current milepost information.This thesis is based on driving images and GPS data point.The main research is a effective and accurate method for mileage pile positioning,achieving the precise positioning of the mileage pile.The subject is based on actual research projects,and the main research contents are as follows:For the different characteristics of kilometer piles and 100-meter piles,the video image oriented algorithms for kilometer pile and 100-meter pile recognition are proposed respectively.For the kilometer piles,firstly dividing connected domain according to the color characteristics,and then selecting the adjusted CRAFT algorithm to detect and recognize the scene text(in the unit of characters)according to the numbering rules of the mileage piles to obtain the pile numbers.Facing the 100-meter pile,given that the 100-meter pile,as a very small target,is difficult to separate based on color characteristics and background.This thesis chooses the its own shape characteristics(Round shape and mounting position)to positioning and segmentation of ROI target detection area.Finally,the Tesseract recognition algorithm is used to digitally identify the area to be recognized.According to the GPS longitude and latitude coordinates and the change curve of the motion law of the detection vehicle,the combination of the CTRV motion model and the UKF algorithm is proposed to achieve effective filtering of GPS information.Since the driving video is GPS information acquired in the time domain,it is necessary to filter and denoise the received GPS data.After comparing the two mature filter methods(EKF and UKF),the CTRV motion model is selected for state transition model of UKF according to the motion law of the detected car,achieving GPS information filtering.Finally,the denoising result is converted into Cartesian coordinate system data and matched with the electronic map to verify the denoising effect.A multi-sensor data fusion algorithm for milepost location is proposed by combining travel video and GPS information.In order to solve the large amount of redundant work generated by the frame-by-frame recognition in the video recognition process,the Markov model is used to improve the mileage pile algorithm detection in the driving video,and a sliding detection algorithm based on speed reference is proposed to reduce meaningless recognition.and then,it can flexibly adjust the recognition algorithm for 100-meter piles and kilometer piles according to the detection time status.pointing at the mileage piles in the video that cannot be entered due to overtaking and other reasons,a mileage pile position interpolation algorithm using Bezier curve fitting partial roads for equal arc length interpolation is proposed to fill in the missing mileage piles improving the accuracy and precision of the data.