Research On RTK/INS/Mono-Vision Fusion Positioning Method For Mass Smartphones

Google Inc.has opened up the Global Navigation Satellite System(GNSS)raw measurements under the Android platform,making it possible to achieve sub-meter level positioning on a low-cost smart mobile platform.However,due to the influence of consumer-grade GNSS chips,linearly polarized antennas,and complex urban observation environment,Android GNSS observations usually contain a large number of outliers,and even with high-precision data processing methods such as Real-time Kinematic(RTK),it is still difficult for smartphones to provide continuous and reliable high-precision positioning.One of the effective ways to solve the above problem is to use the multisource sensors(e.g.,Inertial Measurement Unit(IMU),camera,etc.)of smartphones to achieve multi-sensor fusion positioning.Visual-inertial Odometry(VIO)based on IMU and camera observation information combines Inertial Navigation System(INS)and Monocular-Vision(Mono-Vision)positioning algorithm and it can provide accurate relative position in a short time.However,on low-cost sensors,this method has a fast error accumulation and also suffers from the problem of inaccurate scale recovery.RTK,on the other hand,can provide drift-free absolute position,but the positioning performance will be significantly degraded in complex GNSS observation environments such as urban canyons and boulevards,and the positioning fails in GNSS rejection environments such as indoors.Although RTK and VIO positioning have their own disadvantages,they have good complementarity in terms of positioning modes and environmental applicability.Therefore,the study of RTK/INS/Mono-Vision multi-fusion positioning method for smartphones,which fully exploits the complementary advantages of RTK,inertial and vision sensors,will be of great significance for the improvement of smartphone positioning performance in complex GNSS observation environments.First,a robust RTK positioning method for smartphones based on Factor Graph Optimization(FGO)is conducted in this paper.The FGO framework can be easily extended to incorporate other sensor factors and has advantages in reducing linearization errors and enhancing estimation robustness.To take these advantages,based on the sliding window FGO framework,we proposed a robust RTK positioning scheme that integrates three robust estimation strategies,namely the Cauchy kernel function,Dynamic Covariance Scaling(DCS),and Switchable Constraints(SC),to suppress the effects of outliers,and implements the sliding window marginalization to guarantee real-time performance.In-vehicle positioning experiments are conducted for Xiaomi Mi8 and Huawei P40 smartphones in open-sky and complex urban environments,respectively.The experimental results show that in the open-sky environment,based on the proposed method,the RMS of the positioning errors in the east,north,and up components are 0.18m,0.13 m,and 0.38 m,respectively,providing high-precision positioning;in the complex urban environment,the method combined with the three robust estimation strategies of Cauchy,DCS,and SC robust estimation strategies all effectively suppress the effects of outliers,and the RMS of the horizontal positioning errors using them are 2.38,2.49,and2.43 m,respectively,with the percentage of positioning precision improvement exceeding60%compared with the conventional RTK solution.Then,based on the above FGO-based smartphone robust RTK positioning method,the mathematical model of RTK/INS/Mono-Vision loosely/tightly coupled positioning is established by further introducing the assistance of vision and inertial navigation information.For the problem of blurred images caused by camera shake when recording video under the motion state of smartphone users,digital image stabilization technology is used to assist in video anti-shake,and a GNSS,IMU,and vision-optimized data acquisition software is implemented;for the problem of image distortion caused by the rolling shutter effect of smartphones,the position of feature points is corrected using the frame readout time;for the problem of smartphone IMU-camera time desynchronization error,online estimation is performed by augmenting the state and using visual reprojection constraints;for the problem of a large number of outliers of smartphones in complex observation environments,robust data processing is performed in the preprocessing and nonlinear optimization stages of GNSS and vision.Finally,the proposed positioning method is validated based on the Xiaomi Mi8 smartphone pedestrian experiments in a complex environment.The results show that the RMS of the horizontal positioning error is reduced by 48.0% and 64.2% for the loosely coupled and tightly coupled solutions,respectively,compared to the RTK solution,during the complex GNSS observation time period;and by 36.4% and 58.0%,respectively,to 1.68 and 1.11 m for the loosely coupled and tightly coupled solutions,compared to the RTK solution during the full-time period of the experiment.Therefore,the proposed method can effectively introduce visual and inertial navigation information to improve the robustness of GNSS positioning.The superior positioning performance of the tightly coupled scheme compared with the loosely coupled scheme is mainly attributed to the fact that the GNSS,vision,and inertial navigation observations are modeled and optimized together at once in the tightly coupled framework performed in the observation domain,while the vision and inertial navigation information can assist in the rejection and weighting of GNSS outliers in the observation domain,thus achieving better positioning performance.