Research On Key Technologies Of Lane-Level Positioning And Navigation System For Driverless Vehicles

Positioning and navigation systems are widely used in structured road scenarios as a core component of intelligent transportation systems and driverless vehicles.The key technologies of the positioning and navigation system include three parts:localization to determine the vehicle position,map matching to determine the relative position of the vehicle in the map,and route planning to determine the driving route from the starting point to the end point.When the positioning and navigation system reaches the lane level,it can provide lane level positioning and navigation routes,which in turn provides finer guidance for driverless vehicles and improves the safety and driving efficiency of autonomous driving.The large-scale development and application of driverless vehicles has placed demands on lane level positioning and navigation systems to balance accuracy,stability,cost and efficiency,however,as the structured road environment becomes more complex,there are still some problems to be solved in the research of key technologies for lane level positioning and navigation systems to meet these requirements.In terms of positioning,structured road environments often suffer from GNSS signal occlusion,which affects the positioning accuracy of combined GNSS/INS navigation.Existing low-cost fusion positioning lacks adaptability to multiple driving conditions and positioning stability when using camera-identified lane marking information and wheel odometry to improve GNSS/INS combined navigation accuracy,and the improvement of positioning accuracy is limited.In map matching,Hidden Markov as a common map matching algorithm,it is challenging to improve its matching accuracy and matching efficiency in the interlaced and complex structured road environment at the same time.In terms of route planning,with the increasing scale of structured road networks,how to effectively improve the planning efficiency of existing route planning methods when applied to lane level route search is an urgent issue to be addressed.In this paper,we address the above issues from the perspectives of improving the accuracy,stability and reliability of low-cost positioning and map matching algorithms,as well as improving the planning efficiency of traditional search algorithms when applied to lane level route planning,and carry out the following work:1.Establish an enhanced positioning system based on camera and lane marking,based on the combined positioning of global positioning system(GNSS)and inertial measurement unit(IMU),fusing camera and lightweight lane level map to improve the overall positioning accuracy and stability of the vehicle under driving conditions such as lane change.Firstly,an adaptive multi-indicator weighted evaluation map matching algorithm is proposed in combination with the designed lane change recognition method,and further a lane left and right boundary point determination method is designed;then,on the basis of accurate matching to the lane and its boundary points,a camera-based bilateral boundary line lateral positioning method is proposed to improve the lateral positioning accuracy and stability of the vehicle relative to the lane,and a transition smoothing algorithm is designed to ensure the positioning trajectory is smooth and continuous to meet the needs of the downstream planning and control layer of the driverless vehicle.Finally,real vehicle experiments are conducted to verify the accuracy,stability and continuity of the positioning algorithm.2.Establish a deep learning-based wheel odometry error prediction model for stable and accurate prediction of wheel odometry errors under various driving conditions.Specifically,a Transformer neural network-based wheel odometry error prediction model is designed to learn the uncertainty of wheel odometry measurement and then accurately predict the odometry error.Driving conditions features are considered in the model,specifically features describing road types,road conditions,and vehicle driving operations,and models with and without driving conditions features are compared and analyzed.The prediction accuracy,stability and reliability of the model are verified on a public dataset under various driving conditions,while the model trained on the dataset is transferred to a real vehicle for testing to verify the generalization capability of the model and the improvement of the model on the performance of wheel odometry positioning.3.An improved lane level online map matching algorithm based on Hidden Markov Model(HMM)is proposed.The multi-metric weighted map matching algorithm is used to compensate for the delay of the initial matching of HMM,a mapped hierarchical search algorithm from road to lane is proposed to improve the matching efficiency,a road intersection state switching mechanism is proposed to identify different states,and then a differentiated probability model is proposed to improve the matching accuracy,and a variable sliding window Viterbi algorithm is used to ensure the real-time matching.After that,a robust and low-cost fusion positioning system is designed that fuses GNSS,IMU,wheel odometer,camera and map.The accuracy and efficiency of the improved HMM algorithm are verified through simulation and realvehicle experiments,and the accuracy,stability and reliability of the fused positioning system are verified through real-vehicle experiments.Finally,the map engine,decisionmaking,planning,and control modules of the autonomous driving system are constructed on a simulation platform.The positioning system is integrated with these modules in a joint simulation to verify the effectiveness of the positioning system in supporting the operation of other sub-modules in autonomous driving.4.A lane level route planning algorithm is proposed based on a multi-layer road network model to improve the efficiency of the graph search algorithm when applied to lane level route planning and the adaptability of the algorithm to multiple road network structures.First,a lane level road network model is established that can adapt to the complex road network structure representation,specifically five sub-layers are designed to refine the internal structure of roads and intersection areas so that the model can express multiple variations of road network structure.Then a lane level route planning algorithm is designed based on this road network model that can satisfy traffic rules,vehicle characteristics,optimization objectives and improve the efficiency of the graph search algorithm: firstly,a multi-layer route planning algorithm is designed for hierarchical planning at the road level,lane group level,lane section level and lane level to reduce the search space and improve the planning efficiency.Then an optimal lane determination algorithm considering traffic rules,vehicle characteristics and optimization objectives is developed in the lane layer to be able to find optimal lanes on roads with different structures(including roads with constant or variable number of lanes)while satisfying traffic rules and vehicle characteristics.Finally,the effectiveness of the algorithm is verified on the simulated road network and the real road network.