Lane-level Vehicular Localization Utilizing Smartphones

Accurate vehicle navigation services are highly dependent on accurate road information and exact real-time position of vehicle in the lane level,even for future self-driving cars.A lane-level localization can locate which lane the car is driving along and consider whether to change the lane or not,rather than just knowing where to make turns.This is not only helpful for the drivers going to a strange place,but also a critical technology for navigating self-driving cars in multi-lane roads.As a result,an accurate,real-time,easy popularization lane-level localization system can play an important role in the further development of intelligent transportation.In many existing works,lane-level localization systems use many dedicated and expensive equipments,such as laser radars,which can detect surrounding environments.However,such kind of expensive,dedicated equipments have a negative impact on the popularization of the system.There is another lane-level localization system based on image recognition in recent works,which can identify lanes from images captured by cameras fixed in cars.However,such kind of system can be easily interfered by other factors,such as insufficient lights,road aging,fade road marks,etc.The proliferation of smartphones provides an opportunity for the large-scale lane-level information collection and localization.Today’s smartphone is embedded with a rich set of sensors such as GPS,digital compass and accelerometer,and a smartphone based localization system is easier to popularize.Since the accuracy of GPS is limited and phone cameras consume much power,need to be fixed on cars,and the performance at night is relatively poor.So a system combining acclerometer,compass and GPS is more suitable for complex conditions.Morever,the increasingly popular crowdsourcing is very suitable for large-scale data collection using smartphones.In this paper,we first use IMM algorithm to build two models in order to utilize acceleration data to increase the accuracy of GPS.The proposed constrained K-means algorithm provides lane-level localization service and the way to generate lane-level maps by clustering collected information from users.Moreover,the proposed lane change detection algorithm can detect the action of turns and lane changes,by analyzing orientation and acceleration data.This makes offline localization possible when the initial lane position is known.We also design an incentive mechanism to encourage users to upload their GPS data,and reward them according to the value of their uploaded data.The evaluation results shows the lane-level localization accuracy achieves higher than 95% in our system.