Cloud Positioning System for Intelligent Connected Vehicle in Urban Area

Yue Bingjian ¹ , Shi Shuming ¹ , Yu Jianhua ² , Jia Suhua ¹ , Ma Xiaofan ¹ , Lin Nan ¹

1. Transportation College of Jilin University

2. Advanced Product Development Department of Dongfeng Commercial Vehicle Technology Center

Abstract: Intelligent Connected Vehicles(ICVs)rely on positioning for a number of safety-critical applications.Global Navigation Satellite System(GNSS),alone or in combination with other device,is the most popular positioning system.Due to large noise,multi-path errors and the frequent rise and fall of satellites caused by propagation obstacles and mirror materials,the positioning availability and accuracy based on GNSS are heavily reduced in urban canyons.In vehicular network,lots of location-related measurements are shared by ICVs.The fusion of these multisource measurements is an intelligent strategy for a more robust positioning.In this context,we present a distributed fusion framework,Cloud Positioning System(CPS),to augment the positioning performance.With 5G V2X communications,it is possible to simultaneously provide positioning service for hundreds of ICVs.In the purpose of showing information flow,we assume that the perfect association between vehicles and measurements is available.The core estimation method of CPS is implemented by a Kalman Filter,which provides the upper bound of positioning performance compared to the practical algorithm with data association.CPS availability is analysed by block diagram resulting that:when the availability of GNSS signal decreased from 99% to 90%,the availability of CPS is still higher than 99%.

Key words: cooperative positioning,vehicular network,cloud computing,availability

Introduction

Cooperative Intelligent Transportation System(C-ITS),also known as connected vehicle technology,is a promising technology to enhance driving safety and efficiency.Accurate positioning of traffic participant is a fundamental prerequisite for safety-critical applications in the C-ITS domain ^[1] .Shladover and Tan ^[2] state that 1 m position accuracy could be marginally acceptable for collision warning applications,while 50 cm would yield significantly better results.In general,positioning accuracy is quantified by the distance between measured and true value.Root Mean Squared Error(RMSE),Circular Error Probable(CEP)of 50%(or 95%) ^[3] ,or the ± σ (or ±2 σ )confidence interval are usually used.In addition to accuracy,reliability or availability is another indispensable indicator.Availability means “the ratio of the total time a service is being used during a given interval to the length of the interval.” For example,a service provider may state that its services will be available 99.99 percent of the time over a year,which amounts to 53 minutes of accumulated outages for all causes over the course of the year.In general,the availability of positioning needs to be 100%.

Global Navigation Satellite Systems(GNSSs)have been widely used to provide absolute location and velocity ^[4] .But,even in the open sky areas,standard code-phase-based GNSS position measurements may be biased by as much as 3-5 meters.With sufficient satellites,GPS/IMU(Inertial Measurement Unit)navigation system enables high-accuracy positioning through urban areas ^[5] .With sufficient satellites,visual Simulta neous Localization and Mapping,Carrier-Phase-Based GPS,and IMUs can be fused for obtaining high precision globally-referenced position ^[6] .With sufficient satellites,the fusion of GNSS and digital map may guarantee an accurate of 50 cm ^[7] .Unfortunately,GNSS-based ego-positioning cannot guarantee the availability and accuracy in urban canyons or tunnels,where the satellite signal is severely attenuated or even denied ^[8] .

Cooperative Positioning(CP)is the technique using two or more physical entities work together to improve their positioning with some data fusion method.In the early stages,CP methods use wireless communication devices to share their information and estimate the relative range between broadcasters.[9] and [10]have reviewed and summarized this approach.The augmentation information obtained through V2X communication systems has shown the potential to improve the positioning accuracy and reliability in different traffic environments ^[11] .However,extracting explicit range measurements from the V2X radio signals is suffering from multipath and non-line-of-sight.[12] and [13]present an Implicit Cooperative Positioning(ICP)algorithm avoiding the use of explicit V2V measurements,where ICVs detect a set of non-cooperative physical features in the surrounding areas,and use them as common noisy reference points to refine their location estimates.

The above positioning approaches based on on-board algorithms are named ego-positioning,where the positioning result is named as ego-state.Ego-positioning relies on expensive on-board sensors and complex technical means cost millions of dollars for hardware device and algorithm debug.However,the accuracy and reliability do not meet the requirements of safety-critical applica tions.In addition to ego-positioning function,environment perception and V2X communication are the basic functions of ICV.ICVs are equipped with radar,LiDAR,camera and other perceptual sensors to track surrounding objects.Each ICV can design its own tracker with elaborate algorithm,named ego-tracking,and output the relative state information between ego and neighbor vehicles,named ego-based-tracks.According to the latest standards for C-V2X ^[14][15] ,in most of the communication scenarios,the maximum end-to-end latencies for critical services are given around 20 ms—100 ms with data rates peaking at 1000 Mbps.Through V2X communication,ICVs can share their ego-states and ego-based-tracks to each other with millisecond latency and get road environment information on the RSU,simultaneously.An intelligent strategy is to fuse these multisource measurements for a more robust state estimation.

There is a European project,named AutoNet2030,connecting two domains of intensive research:C-ITS and ICV,which is committed to studying the connection,communication and organization ^[16] .A cloud-assisted system for autonomous driving named Carcel is presented in [17],which is used to avoid obstacles and plan efficient paths that account for unexpected events.A centralized cloud service is used in [18],which is in charge of receiving,merging,and distributing data acquired by different sensing systems for implementing advanced navigation strategies.When multiple ICVs gather in a small area,they can locate themselves and track other ICVs,so as to measure a certain ICV in many times.Based on information redundancy caused by the multi-measurement,it is certain that the fusion of these measurements can further improve positioning accuracy.

In this paper,we propose a distributed fusion framework,Cloud Positioning System(CPS),for vehicle positioning improvement.ICVs are information sources provided CPS with multi-target state and ITS-station is a fusion center deployed to send estimated state to each ICV.Thanks to the high transmission rate,precise clock,and network timing of 5G C-V2X,the measurement synchronization can be solved by a predict step.In the purpose of showing information flow,we do not explicitly treat Data Association(DA)problem,but rather assume that the perfect association between vehicles and measurements is available.Based on this,the proposed CPS can be implemented by a Kalman Filter,which provides the upper bound of positioning performance compared to the practical algorithm with DA.We use block diagram to analyze the availability of CPS at different availability levels of GNSS,which shows the CPS can provide continuous positioning service for ICVs driving in urban environments.

The rest of the paper is organized as follows.In Section 1,we describe the CPS framework and potential service capacity.The implementation is shown in Section 2.Section 3 introduces the benefit on availability.Finally,Section 4 concludes this paper.

1 Cloud Positioning System for Intellgent Connected Vehicle

In this section,we describe the CPS in detail.Firstly,the ideal C-ITS scenario model is introduced in Section 1.1.Then,in Section 1.2,we introduce the CPS framework.At last,the potential service capacity of our CPS is assessed in Section 1.3.The fusion approach will be discussed in Section 2.

1.1 Ideal C-ITS Model

In the ideal C-ITS scenario,every vehicle is an ICV with three functions:connect to the C-ITS network by V2X,estimate its own state,and track different objects in surrounding environment.As shown in Fig.1,the red car is locating itself by GNSS,and it is tracking objects within a sensing range at the same time.Here,we assume that the ego-positioning is in 2D global coordinate and ego-tracking is formatted with relative position and velocity in vehicle coordinate.Furthermore,the positioning and tracking errors are formed of White Gaussian Noise(WGN).

As an example of ideal C-ITS scenario,we consider 4 ICVs driving in the same direction,as shown in Fig.2.For each vehicle n ,the dynamic state is characterized by

where [ p _{x , n} , p _{y , n} ] ^T ∈R ^2×1 is position and [ v _{x , n} , v _{y , n} ] ^T ∈R ^2×1 is velocity.Furthermore,we assume the dynamic state evolves over time k according to a linear Gaussian model

where denotes the transition matrix; T _s is the sampling interval; and q _n denotes the process noise. q _n is modeled as zero mean Gaussian with known covariance Q ∈R ^4×4 , i. e., q _n R N ( 0 , Q ).

The ego-state estimated by on-board sensors and algorithm is formatted as

where r _n ～ N ( 0 , R ) denotes the measurement noise.Moreover,vehicle n can also calculate M ego-based-tracks generated from surrounding objects with

where s _m is the dynamic state of target object associated to measurement T _{n , m} ,and n _{n , m} ～ N ( 0 , N _{n , m} ) . It can be seen that, T _{n , m} is equal to the difference between target state s _m and ego state s _n .Here,we assume that each target object s _m can generate at most one track within the sensing range.

Take the red car in Fig.2 as an example,if it is vehicle n =1,the ego-state L ₁ and ego-based-tracks T _1,1 ～ T _1,3 can be sent to ITS-station through V2X. Similar to the red car,the other three cars will also send their own ego-state and ego-based-tracks,simultaneously.

1.2 Cloud Positioning System Framework

As shown in Fig.3,the CPS proposed in this paper is a distributed fusion framework based on ideal C-ITS scenario.Each ICV is regarded as a mobile sensing node.So,as the end of vehicular networks,ICVs are regarded as mobile sensing nodes,where the raw on-board sensor data are processed by ICV itself resulting ICV's state(ego-state)and local multi-target track(ego-based-tracks).These ego-state and ego-based-tracks are sent to ITS-station through V2X devices.The ITS-station deployed with advanced hardware is the fusion center of CPS,where data reconciliation,association,fusion,and feedback are implemented.The ITS-station will complete global estimation of each node and send the fusion results,such as ICV's state,back to each ICV through V2X.

The fusion of CPS can be treated as Multisensor-Multitarget Tracking(MMT)problem,which involves the joint estimation of an unknown and time-varying number of targets as well as their individual states from a sequence of sets of noisy measurements from different sensors.If we get the solution of this MMT problem,as part of all objects,the positioning accuracy of ICVs will be improved by multiple measurements.The tracks output by ITS-station are the temporal sequences of estimated object states.Each ICV must be associated with at least one track for closed loop feedback of state estimation.The approach for this MMT problem and state estimation feedback under perfect data association is in Section 2.

1.3 Potential Service Capacity

Our proposed positioning approach requires significant communication between ICV and ITS-station.At each time step,ego-state L _n =[ p _{x , n} , p _{y , n} , v _{x , n} , v _{y , n} ] and ego-based-tracks Tr _n = are sent by each ICV.So,the total number of messages per vehicle is more than (1+ M ),where M denotes the number of ego-based-tracks.Considering a data rate of R bits/s and expected communication period t _s ,the number of ICV connected to CPS is upper bounded by:

where N _bit is the number of bits needed to describe an ego-state or ego-track.

As an example,in the case of using the 3GPP LTE C-V2X,with R =50Mbit/s, M =24ego-based-tracks, N _bit =200 bits(each message contains at least 6 single-precision floating-point numbers,occupying 32×6 bits,for time,ID,position,and velocity),and t _s =0.1 s,we find that N _ICV =1000,which is a considerable and satisfactory number.

2 Implementation of Fusion Center

In this section,we will introduce the fusion approach of ITS-station in detail.First,the data integration method is introduced in Section 2.1,which mainly solves the problem of measurement synchronization.Then,the DA problem that CPS needs to solve is introduced in section 2.2.The Kalman Filter is used as data fusion method in Section 2.3 under the assumption of perfect association,which provides the upper bound of positioning performance compared to the practical algorithm with DA.

2.1 Measurement Synchronization

The proposed CPS approach is built on the assumption that all measurements are taken and processed synchronously.However,rarely ego-positioning and ego-tracking at a given vehicle and across different vehicles are isochronous.At each time step,temporal realignment of messages is mandatory and could be handled as follows.Thanks for 5G timing control,ICV clocks are well synchronized according to vehicular network clock.Considering the latency requirement of safety-critical ITS applications,CPS,as any other cooperative positioning approach,is expected to rely on very-low end-to-end communication latency.Within 5G modes,which is the new-generation V2X systems supporting connected automated driving and devoted to vehicular communications,a latency of 1-10 ms is envisioned.If the time elapsed between sensing and transmission is large enough,there is a prediction step for synchronization,typically similar to the Kalman filter based on dynamic models.Assuming that CPS can rely on very-low V2V latency and on a temporal re-alignment of measurements as above discussed,measurements can be re-synchronized to a common time horizon at each time step,with the price of an increasing variance.Clearly,as the time difference between measurements are increased,data fusion is less accurate.Furthermore,this imposes to restrict the time can be used in ITS-station.

2.2 Data Association

A certain form of DA is required for the implementation of our CPS approach.In particular,there are two data associations for cooperative positioning.The first is all measurements collected by ITS-station need to be classified into N clusters,where N is expected to the number of vehicles in the scenario shown in Figure 2.As a specific example in,the ego-state L ₃ and ego-based-tracks { T _4,1 , T _1,2 , T _2,3 } must be matched as one cluster.The second is the N clusters should be matched with corresponding vehicles,otherwise,the estimated state cannot be fed back for the improvement of positioning.Once these DA problems are solved,the measurements of a certain vehicle can be combined to get an enhanced positioning.

Most traditional DA method,such as Multiple Hypotheses Tracking(MHT) ^[19] ,joint probabilistic data association filter(JPDAF) ^[20] ,and the Multitarget particle filter ^[21] ,involve explicit associations between measurements and vehicles.Due to its combinatorial nature,DA problem makes up the bulk of the computational burden.The main alternative formulation that avoids explicit associations between measurements and targets is Random Finite Sets(RFS),which is also known as Finite Set Statistics Theory(FISST) ^[22,23,24] .

ICV can track hundreds of targets,whereas most valuable tracks should be selected by considering:(a)well separated in either distance,angle or velocity are preferred,as they are easier for DA; (b)attribute information,such as class(car,bicycles,and so on),color,and dimension,to ease DA.The latter various discrete state information may be an important approach to optimize DA performance and reduce computational burden.

In this work we do not explicitly treat DA problem,but rather assume that the perfect data association between ICV and measurements is done,based on which the proposed CPS can be implemented.In this context,the Kalman Filter provides the upper of positioning performance compared to the practical algorithm with DA.

2.3 Data Fusion under Perfect Association

In our work,ITS-station is a fusion center with global coordinate.All ego-states and ego-based-tracks can be shared to ITS-station for cooperative positioning.Before the fusion of all data,we need solve the problem of inconsistent coordinates,where ego-based-tracks are in vehicle coordinate.

Because the ego-state L _n is measured in global coordinate.So,we directly use it as the ITS-station measurement,that is

For ego-track T _{n , m} ,we get the following conclusions according to(3)and(4),

where r _{n , m} = r _n + n _{n , m} .Because ego-state and ego-based-tracks are implemented by different sensors and algorithms,we assume that r _n and n _{n , m} are independent.So,the overlaid noise r _{n , m} .is still WGN with covariance matrix R _{n , m} = R _n + N _{n , m} .

Finally,we get the global coordinate expression.

where the subscript of 100 n +0 and 100 n + m helps to avoid duplicate numbers. As shown in Fig. 4, the ego-state L ₁ and ego-based-tracks T _1,1 ▯ T _1,3 are converted to ITS-station measurements z ₁₀₀ , z ₁₀₁ , z ₁₀₂ and z ₁₀₃ .

Under the perfect data association, z _j generated by vehicle n is selected to compose measurements set . Each measurement in z _j in set . is generated by different sensor and algorithm and can be modeled as

where H = I ₄ is the measurement matrix,and R _{n , k} is the measurement noise belonging to WGN.

Taking the green car in Fig.2 as an example,as we know the association between vehicles and measurements,there are 4 measurements z ₃₀₀ , z ₄₀₁ , z ₁₀₂ ,and z ₂₀₃ (that is L ₃ , T _4,1 , T _1,2 and T _2,3 )can be used to estimate a more accurate location.The tracks output by ITS-station are the temporal sequences of esti mated object states.Each ICV must be associated with at least one track for closed loop feedback of state estimation.Then,the estimated state x _{k | k} with a more positioning accuracy can be fed back through V2X.Based on Bayesian filtering and our assumptions,dynamic model and measurement model are linear Gaussian.The state estimation in CPS,both in ICV and Fusion Center,can be reduced to the Kalman Filter.From the perspective of ICV,Sequential Kalman Filter can be used to process multiple measurements,that is,to perform multiple correction steps.The final estimated state is the result of the last correction,which will be used in the prediction step of next iteration.Still taking the green car in Figure 2 as an example,Fig.5 shows this process.

3 CPS Availability

In urban canyons or tunnels,the number of GNSS satellites is greatly reduced because they are severely blocked by tall build ings and trees.If the satellites can be observed is lower than requirements,ego-positioning algorithm based on GNSS is unavailable in the confined environment.The availability of a satellite navigation system is the percentage of time that the navigation sys tem can provide a vehicle with usable navigation services in service space ^[25] .

In reliability domain,steady-state availability is an index used to evaluate system availability and defined as

where MTBF is the mean time between failures,MTTR is the mean time to repair, λ is the failure rate,and μ is the maintenance rate.If the availability of an equivalent unit is known to be A _i ,it can be known

Fig.6 shows the availability block diagram of a certain ICV in CPS,which is simplified into three parts.The first part is a parallel system composed of multiple ICVs with availability A _Para ,where we assume the availability of sensor is A _S ,the availability of ego-positioning is A _P ,and the availability of ego-tracking is A _T .The second part is a serial V2X communication with availability A _C .The third part is the series ITS-station with availability A _F .

According to the calculation formula for the availability repairable series system,the availability of IVC is

Assuming that all ICVs have the same availability,according to the calculation formula for the availability of homogeneous parallel system,the availability A _Para of N ICVs is

So,the availability of CPS is

As a more specific example,Tab.1 shows the availability of CPS system composed of 4 ICVs when A _S is at multiple levels.Assuming that the availability of ITS-station,which is determined by hardware device and fusion method,is A _F =100%.The availability of 5G C-V2X communication in urban environment is A _C =100%.The availability of on-board methods are A _P =100% and A _T =100%.

It can be seen that the availability of ego-positioning, A _Ego is proportional to and decreases rapidly with A _S .Thanks to the redundant measurement of parallel system, A _CPS is 1%-19% higher then A _Ego .When A _S drops to 90%, A _CPS is still higher than 99%.

According to the category of GNSS,some vehicles maintain the availability and accuracy of ego-positioning,whereas other vehicles in the same area cannot connect to satellites.When multiple ICVs estimate their locations and track other ICVs in a small area,redundant measurements which are perfect associated to a certain vehicle can be fused to improve positioning availability.

4 Conclusion

In this paper,firstly,a novel CPS for ICV cooperative posi tioning is proposed.The CPS resulting positioning improvement is based on distributed framework and the fusion of multisensor multitarget tracking data.Secondly,system availability analysis results show that:when the availability of GNSS in an urban area is decreased from 99% to 90%,the availability of CPS is still higher than 99%.Thirdly,some implementation aspects related to communication burden,latency,and measurement synchronization are discussed for the feasibility of CPS.

There are still several problems need to be solved,such as DA problem,decision-level fusion algorithm,and the assumption of point target model.In addition,the performance needs to be evaluated in more complex scenarios.Under the CPS framework,the vehicle positioning is transformed into MMT problem which a variety of proved effective methods can be used.We are exploring the way to achieve these goals and some solutions are undergoing peer review.

This work was supported by the National Natural Science Foundation of China under Grant U1964202.

References

[1]CAVENEY D.Cooperative Vehicular Safety Applications[M].New Jersey:John Wiley & Sons,2009.

[2]SHLADOVER S,T S K.Analysis of Vehicle Positioning Accuracy Requirements for Communication-Based Cooperative Collision Warning[J].Journal of Intelligent Transportation Systems,2006,10(3):131-140.

[3]PINTO N J,GOMES L C,ORTIZ F M.Almeida,Thales T.; Campista,Miguel Elias M.; Costa,Luís Henrique M.K.; Mitton,Nathalie.An accurate cooperative positioning system for vehicular safety applications[J].Computers & Electrical Engineering,2020,83:106591.

[4]BONET B,ALCANTARILLA I,FLAMENT D et al.The Benefits of Multi-constellation GNSS:Reaching up Even to Single Constellation GNSS Users[C]//International Technical Meeting of Satellite Division of The Institute of Navigation(ION GNSS 2009),2009.

[5]KENNEDY S,HAMILTION J, MARTELL H. Architecture and system performance of SPAN NovAtel′s GPS/INS solution[C]//2006 IEEE Position Location and Navigation Symposium,2006,pp.266-274.

[6]SHEPARD,DP,HUMPHREYS,TE.High-Precision Globally-Referenced Position and Attitude via a Fusion of Visual SLAM,Carrier-Phase-Based GPS,and Inertial Measurements[C]//2014 IEEE -ION Position Location and Navigation Symposium,2014,pp.1309-1328.

[7]ALAM N,DEMPSTER A G.Cooperative positioning for vehicular networks:Facts and future(Article)[J].IEEE Transactions on Intelligent Transportation Systems,2018,14(4):1708-1717.

[8]KAPLAN E D,HEGARTY,CJ.Understanding GPS:Principles and Applications Second Edition[M].Fitchburg:Artech House,2006.

[9]NIMA A,ANDREW G D.Cooperative Positioning for Vehicular Networks:Facts and Future[J].IEEE Transactions on Intelligent Transportation Systems,2013,14(4):1708-1717.

[10]SAMPO K, SABER F, KONSTANTINOS K, et al.A survey of the state-of-the-art localization techniques and their potentials for autonomous vehicle applications[J].IEEE Internet of Things Journal,2018,5(2):829-846.

[11]MARCUS O, NORMAN M, ROBIN S, et al.Car-to-Car communication for accurate vehicle localization — The CoVeL approach[C]//International Multi-Conference on Systems,Signals & Devices,2012,pp.1-6.

[12]SOATTI G,NICOLI M,GARCIA N,et al.Implicit cooperative positioning in vehicular networks[J].IEEE Transactions on Intelligent Transportation Systems,2018,19(12):3964-3980.

[13]MATTIA B,MONICA N,GLORIA S,et al.Augmenting Vehicle Localization by Cooperative Sensing of the Driving Environment:Insight on Data Association in Urban Traffic Scenarios[J].IEEE Transactions on Intelligent Transportation Systems,2020,21(4):1646-1663.

[14]3GPP.Study on enhancement of 3GPP Support for 5G V2X Services[R].3GPP TSG SA1 TR 22.886,2017.

[15]3GPP.Study on LTE Support for Vehicle-to-Everything(V2X)Services[R].3GPP TR 22.885,2015.

[16]ARNAUD DE LA F,XIANGJUN Q,SEBASTIEN D,et al.Network of automated vehicles:the AutoNet 2030 vision[C]//ITS World Congress,2014.

[17]SWARUN K, SHYAMNATH G, DINA K.A cloud-assisted design for autonomous driving[C]//MCC′12:Proceedings of the first edition of the MCC workshop on Mobile cloud computing,2012.

[18]CARDARELLI E,SABATTINI L,SECCHI C,et al.Cloud Robotics Paradigm for Enhanced Navigation of Autonomous Vehicles in Real World Industrial Applications[C]//2015 IEEE International Workshop on Intelligent Robots and Systems,2015,pp.4518-4523

[19]STEFANO P C,CRAIG A C.Multiple-hypothesis tracking for targets producing multiple measurements[J].IEEE Transactions on Aerospace and Electronic Systems,2018,54(3):1485-1498.

[20]THOMAS E F,YAAKOV B S,MOLLY S.Sonar tracking of multiple targets using joint probabilistic data association[J].IEEE Journal of Oceanic Engineering,1983,8(3):173-184.

[21]CARINE H,J L C,PATRICK P.Sequential Monte Carlo methods for multiple target tracking and data fusion[J].IEEE Transactions on Signal Processing,2002,50(2):309-325.

[22]Ronald P.S.Advances in Statistical Multisource-Multiple-target Information Fusion[M].Norwood:Artech House,2014.

[23]RONALD P S.Statistical Multisource-multiple-target Information Fusion[M].Norwood:Artech House,2007.

[24]RONALD P S.Random Sets in Information Fusion an Overview[J].Random Sets,1997,97:129-164.

[25]LI Z H.Research on Monitoring and Assessment of Satellite Navigation System Performance[D].Zhengzhou:Information Engineering University of the People's Liberation Army,2012. I9Tqoh7rafq//woQvJ73TmSHNtGYtAH+i6RIOZjf4envUATNX7mJvLsvX+oJKm3p