Traffic State Recognition Based On Floating Car Data Spatial Semantic

Congestion is one of the most serious problems in the field of transportation in China.To alleviate this problem,floating car monitoring and dispatching platforms have been established in many cities.The floating car system has advantages of low cost,high coverage rate and strong real-time performance,which can provide real-time feedback of traffic information.The establishment of the system can alleviates traffic congestion to a certain extent.In this paper,we proposed a temporal and spatial semantic model of FCD(Floating Car Data)for the traffic state identification,which is divided into three parts: spatial-temporal location semantics,road geometry semantics and road scene semantics through in-depth mining and semantic analysis of FCD.Meanwhile,for each part,a corresponding algorithm was proposed to solve the existing problems.The main research work and contributions are reflected in the following aspects:(1)A floating vehicle blind spot location coding algorithm based on monocular vision was proposed.The floating vehicle in the blind area of GPS cannot get the positioning data through GPS.To solve this problem,a spatio-temporal semantic coding method based on monocular vision was proposed.In this method,automobile data recorder is used to obtain the image data around the vehicle.The distance between 2 frames of image is obtained by combining with the wheel tachometer.The 3D reconstruction technology of road scene with monocular vision is used to construct perspective n-point(Pn P)problem.The position of floating car can be calculated according to the position relation of adjacent 2 frames.Compared with other blind location methods,the proposed algorithm based on monocular vision has the advantages of low cost,simple and easy to popularize.Experiment results show that the average accuracy error of the position data obtained by the proposed method is 8m,lower than the average error of GPS,which can be used as a means to obtain the position data of the floating vehicle in the blind area.(2)A geometric semantic coding algorithm based on curve local matching was proposed.In various applications based on floating vehicle data,the road where FCD is located and its corresponding geometry and connectivity are the key issues.In this paper,the geometric shapes of roads and the connectivity between roads are defined as road geometric semantics,and a road geometric semantic coding algorithm based on partial curve matching was proposed.This algorithm reconstructs road network topology based on FCD to obtain the geometric shapes and interconnectivity of roads where each FCD data resides.Fréchet distance was used as the criterion to measure the similarity of curves,the concept of free space of curves is introduced to realize the local matching of curves and graphs,and the similarity judgment of curves is transformed into the problem of monotone shortest path in free space,which reduces the time complexity of the algorithm.Experimental results show that the proposed algorithm is superior to the classical method in terms of geometry and connectivity of roads.(3)A semantic coding method of road scene based on deep learning network is proposed.According to the image data provided by floating vehicle dash-recorder,floating vehicle data are classified through road scenes semantic analysis.Deep learning technology is applied to the global feature extraction of road scenes,and Road Net is built for the semantic classification of road scenes,so as to establish the semantic coding model of road scenes based on deep learning.Experimental results show that the proposed method can effectively and accurately classify road scenes.Compared with other networks,the deep learning network Road Net has higher accuracy and robustness in road scene classification.(4)A road network traffic state identification method based on spatio-temporal semantic coding was proposed.Based on the constructed spatio-temporal location semantics,road geometry semantics and road scene semantics,the semantic calibration of road network traffic state is carried out.Based on GPS data and image data,a method to identify road network traffic status is proposed.A number of trunk roads in Wuchang district of Wuhan were taken as examples to verify and identify the road traffic status.Our research can expand the application of floating car technology.The proposed methods can improve the range and efficiency of identifying traffic conditions based on floating car.The research results can provide a technical and data basis for easing or even improving the current situation of road traffic congestion.