| Intelligent and Connected Vehicles(ICVs)can avoid traffic accidents caused by human driver's misoperation or illegal operation,which is expected to provide safer travel services and even achieve zero road traffic injuries.So,it is an internationally recognized future development direction and focus of the automobile industry.The accurate location information of ICV is the basic prerequisite for various safety functions.However,the existing on-board positioning methods,the traditional satellite positioning enhancement methods,and the Cooperative Positioning(CP)methods ranging by radio signal cannot satisfy positioning accuracy,reliability,and cost at the same time.In the context of the continuous improvement of the ICV's on-board positioning,the greatly improved environmental perception and the multiplication of the Vehicleto-Everything(V2X)communication the CP method using the detections of the onboard target detect sensor has become a new filed of the modern CP.At present,researchers have initially formed a consensus on the use of Multi-Target Tracking(MTT)methods to achieve this goal,but there is no mature technical route has been formed.This filed is a blue ocean that needs to be explored.Based on the in-depth investigation of the latest domestic and foreign researches,considering the internal relationship of vehicle kinematics model,on-board positioning method and on-board sensors' characteristics,this article carries out the modeling of multi-vehicle CP system,the research of MTT method,and the simulation verification method of urban complex traffic scenario.Based on the theory of multi-source and multi-target information fusion,this paper explores the concept and method of cloud positioning system that integrates the results of mutual observation between vehicles under the condition of unknown matching relationship between the target and the observation.this paper provides a new idea and some new methods for low-cost,highprecision and high-reliability ICV positioning.This work is a free exploration of novel and cutting-edge research directions,which is quite challenging.The main research work includes the following 5 aspects.1)The ICVs' cloud positioningapproach with technically feasibilityThis paper studies the relationship between the on-board positioning results and the target detection results in a global perspective,and proposes a cloud positioning concept that uses the detection results to perform CP under the condition that the matches between the targets and the observations are unknown.In short,cloud positioning idea is ‘converting the relative vector of vehicles and objects into global observations,thereby forming vehicle multi-observation data,and using multiobservations to improve vehicle positioning.' In order to verify the implementation conditions of the cloud positioning approach,the Apollo D-KIT Lite intelligent vehicle is used as the data collection platform,and multiple sets of vehicle experiments are designed and carried out.According to the experimental data of Car-Following,the condition that ICVs can obtain the target's global location is verified,and the possibility of implement under 4G or 5G communication conditions is discussed.Then,this conclusion that the cloud concept has technically feasibility and can be implemented when5 G communication is proved.2)The system design and mechanism analysis of the cloud positioning frameworkRelying on typical intelligent transportation scenarios,a cloud positioning system framework is designed based on a distributed fusion structure.By taking the result of single-vehicle positioning as the target state and transforming the relative vehicleobject vector into a global observation,the multi-vehicle cooperative positioning is modeled as a multi-target tracking problem with known target numbers and priori states,and then a complete cloud positioning system model is constructed.Focusing on the requirements of different on-board positioning methods for feedback,the two feedback mechanisms of 'Virtual GNSS' and 'State Deep Fusion' are discussed,which form a closed-loop mechanism to improve vehicle positioning performance.Under the assumption of perfect data association,the Kalman filter is taken as an example to analyze the benefits of multiple observations,and then the reason why the cloud positioning system can improve positioning is revealed.According to the theoretical formula and the reliability block diagram,the influence of the observation number on the positioning accuracy and reliability are analyzed.3)The ellipsoid gate ICV cloud positioning method for highway environmentFocusing on the characteristics of low clutter and low vehicle density in open environments such as highways,under the guidance of traditional MTT theory,a cloud multi-observation fusion method based on the ellipsoid gate rule and sequential Kalman filter is proposed.A simulation experiment system including on-board positioning,target detection,and cloud positioning method is constructed.The positioning accuracy,reliability,and real-time are analyzed through the RMSE of location errors,the CDF of loction errors,and the CPU time of computer.The results show that the ellipsoid gate cloud positioning method has an improvement of 28.75% in median error,an improvement of 25.00% in the worst case,an improvement of 16.69 percentage in the reliability of sub-meter-level positioning.And this method also has excellent real-time performance.4)The Gaussian Mixture Probability Hypothesis Density(GM-PHD)cloudpositioning method for urban environmentIn order to make the cloud positioning system adapt to the dense multi-target urban environment with clutter and missed detection,combined with the characteristics of the ICV actively connecting to the network and uploading vehicle status information and object detections,a Gaussian Mixture Probability Hypothesis Density(GM-PHD)cloud positioning method based on the random finite set theory is proposed.Considering the influence of the urban environment on the sensors,the performance of the two cloud positioning methods is compared and analyzed by adjusting the parameters of the high-speed overtaking scenario and simulating clutter and missed detection.The simulation results show that the GM-PHD cloud positioning method can improve the positioning performance of ICVs in a noisy environment.The median positioning error is improved by 26.79%,the worst-case error is improved by 26.62%,and the submeter position reliability is improved by 15.33 percentage compared with the on-board positioning.And the average CPU time is 35.06 ms,which can meet the needs of 10 Hz positioning service.5)The performance of the cloud positioning methods are verified in the virtualcomplex traffic scenarioIn order to test the cloud positioning methods in urban traffic scenario,microscopic traffic simulation technology is used to reproduce the complex road network structure and diverse driving behaviors in the urban.A street test scenario is established.The one-step prediction is used to determine the simulation parameters such as the process noise covariance matrix.The two cloud positioning methods are verified and analyzed under the condition of with/without "clutter and noise".The simulation results show that the GM-PHD method can effectively improve the positioning performance of ICVs in urban environments,where the median error is improved by 12.08%,the worst-case error is improved by 8.30%,the reliability of submeter-level positioning is improved by 6.66 percentage.The CPU time is 51.99 ms in averaged.The GM-PHD method is proved that they can provide high-precision positioning at 10 Hz... |
PDF Full Text Request