| The rapid development of technologies such as autonomous vehicles and Unmanned Aerial Drones have put forward higher requirements for positioning and locationing.As the main application scenarios of smart vehicles,the complex and diverse characteristics of the building structure and physical environment in the city not only seriously interfere with the quality of the observation signal,but also have an important impact on indicators such as positioning accuracy and stability.The complex signal blockage in urban areas has become a "bottleneck" problem that restricts high-accuracy positioning in urban environments.This dissertation conducts a deep research on the high-accuracy tightly-coupled GNSS/INS integrated positioning technology and its performance evaluation methods in the urban environment.Aiming at the prominent contradiction between autonomous navigation,safety control,and high-accuracy positioning availability of vehicles in complex scenes such as cities,the mechanism on how the scene characteristics affect the accuracy,real-time performance,and reliability of tightly-coupled GNSS/INS integrated positioning is studied.A set of improved tightly-coupled positioning methods were proposed,based on multi-system multi-frequency measurement and machine learning methods,which realize the active adjustment of the measurement model according to the scene characteristics.The contributions effectively improved the accuracy,reliability,and continuity of tightly-coupled positioning in complex urban scenes.The main work and contributions of the thesis are as follows:(1)The tightly-coupled GNSS/INS positioning principle was explained,the nonlinear fusion filter is introduced.According to different state parameters,a tightly-coupled integrated positioning state model and a measurement model were compared.The system noise modeling based on the random walk and first-order Markov chain was analyzed.The GNSS signal transmission characteristics affected by blockages in an urban complex environment were analyzed.By simulating the limitation of the satellite azimuth and elevation angle by buildings,a simulated urban canyon model was designed to verify the positioning performance of the algorithm under different blockage conditions.(2)Utilizing the advantages of EWL/WL short-baseline single-epoch ambiguity fixing,the measurement model was designed with the multi-frequency WL measurement with stepwise fixed ambiguity resolution instead of the traditional carrier or pseudo-range measurement.The improved GNSS/INS integrated measurement model based on multi-frequency information was so that proposed.As the test result,the proposed algorithm based on multi-frequency has achieved a stable decimeter-level positioning capability with a horizontal accuracy of 0.152 m and vertical accuracy of 0.196 m.After 10 s and 30 s signal interruption,compared with the conventional algorithm with multi-epoch ambiguity fixing,the proposed method achieved to increase the positioning accuracy by 44.1% and 28.5%.(3)Aiming at the problem of insufficient observable satellites in the urban environment,a tightly-coupled GNSS/INS integrated positioning method developed with the inter-system differential model was proposed.By introducing inter-system error estimation,the use of a global pivot satellite across all GNSS systems to form double-difference observations improved the positioning accuracy with more redundant observations and better geometry distribution.The verification with simulated urban canyon proved that the fewer satellites observed,the greater improvement the proposed algorithm performed.When testing in a real urban canyon,an accuracy of 49.7cm in horizontal and 67.0cm in vertical were realized with the proposed algorithm.(4)For the difficulty of modeling complex variated measurement noise,a machine learning assisted measurement noise model for tightly-coupled GNSS/INS integration was proposed to realize high-accuracy positioning with adaptively adjusting measurement noise with environmental variation.The strong environmental correlated parameters such as multifrequency SNR,azimuth,and elevation angle were analyzed and then used as features for model training,the measurement residuals were used as the label.The GRU was used to train the model and predict the measurement noise,with labels formed with offline processed groundtruth values.The offline training and online regression strategy avoided problems such as label obtaining difficulty during online training,high computational pressure,and the lacking sample data for training.As the actual test results of complex conditions presented that,the proposed adaptive modeling with machine learning achieved an accuracy of horizontal 14.02% and vertical 20.49% improved compared to the conventional empirical modeling,which effectively improved the accuracy and stability of tightly-coupled integration in an urban environment.(5)Based on the models and methods proposed in the thesis,a tightly-coupled GNSS/INS integrated positioning system was developed for improving the positioning performance in urban areas,which achieved dynamic continuous decimeter/sub-decimeter –level vehicular positioning performance in real application environments.The hardware of the self-developed system was composed of MEMS INS and a receiver that supported multi-system multifrequency observation.The software was developed based on a simplified micro-kernel architecture on basis of the team research contributions and the methods specifically for urban areas.The performance verification of the self-developed positioning system was carried out through the reference system with zero-baseline homology observation and real test datasets.The result showed that the horizontal and vertical positioning accuracy of the self-developed system reached 0.140 m and 0.272 m respectively,which proved the dynamic positioning capabilities at the decimeter level in an urban environment. |
PDF Full Text Request