| As an essential national infrastructure,the Global Navigation Satellite System is capable of positioning,navigating and timing globally,precisely and continuously.However,the GNSS signals suffer from blocking,jamming and spoofing issues when arrived the earth ground,making it impossible to be applied in electromagnetic interference and physical shielding environment.In order to guarantee the robustness of the national PNT system and improve the ability of navigation and location services,the United States proposed a plan named All Source Positioning and Navigation.In the meantime,China has launched ‘Xihe' System and developed the Precise Cooperative Realtime Positioning technology,immediately after which the construction of national comprehensive PNT system with the core of Beidou is initiated.All these projects view multi-sensor integration and multi-source heterogeneous information fusion as their development orientation,which is also an effective approach to conquer the limitation and vulnerability of single navigation system.With the advent of the intelligent navigation era,the industry applications represented by mobile survey and the mass market applications represented by location services have a huge demand for precision positioning and positioning technology.Mobile survey with satellite-based,airborne-based,vehicle-based and ship-based platform,as well as the autonomous navigation of smart carrier such as self-driving car,Unmanned Aerial Vehicle,relies heavily on precise location-based information.Therefore,the precise position and attitude determination technology based on multi-sensor fusion has a promising research significance.This paper aims to intensely study the theoretical model and technical methodology of GNSS/SINS/Vision fusion improving its ability of precise position and attitude determination under complex environments.The main work and contribution of this paper is as follows:1)From the perspective of model simplification and unification,the loose/tight coupled integration model of precise single point positioning and differential positioning is summarized,and the multi-source constraint information and fusion strategy of ground vehicles is discussed.Moreover,for the vehicle scene,a cascade alignment scheme mainly based on speed is proposed.Apart from that,a forward/backward mechanism algorithm is presented in order to achieve bidirectional filtering and bidirectional smoothing.In addition,in the dual-antenna GNSS/SINS combined attitude deamination,the misalignment angle model is employed instead of the Euler angle model to achieve compatibility with the loosely coupled integration.2)A new structure named “semi-tight coupled integration” is designed,which not only enjoys the advantages of loose coupled integration and tight coupled integration,but also overcomes their main defects.It can maintain the same position and attitude determination accuracy as the tight coupled integration in the case of insufficient satellites.and eliminates the risk of ambiguity fixing in "hold" mode of tight coupled integration.Moreover,the RTS smoothing of the fixed solution can also be realized.The novel structure is an effective method suitable for multi-sensor distributed filtering.3)In order to enhance the ambiguity fixing performance in complex environments,three new methods are proposed from the perspectives of positional domain,atmospheric domain and ambiguity domain,namely INS-aided PPP ambiguity fixing,S2L-RTK with ionospheric modeling constraints and Ambiguity Domain Based Integration.Theoretical analysis and field test data process show that when the positioning accuracy of INS dead reckoning is better than half of the wavelength,it can assist the ambiguity to achieve instantaneous ambiguity fixing.Compared with the weighted ionospheric RTK model,S2L-RTK predicts the constraint by ionosphere modeling,by means of which the fixed rate of ambiguity in complex environment is improved by nearly 50%;the post-processing algorithm of ambiguity integration can transfer the correctly fixed ambiguity to the whole arc segment,and realize the ambiguity fixing for the whole arc segment.4)A new GNSS/SINS post-processing strategy is proposed.The strategy uses two independent threads to perform forward/backward Kalman filtering.After filtering completed,each of them performs RTS smoothing individually.Finally,FBC combining technique is applied to forward/backward smoothing results to obtain high precision results.Further,by means of state dimension reduction and update rate decreasing,the post-processing efficiency is greatly improved without loss of precision.It only takes 4.5s to process the data with a length of 2.65 hours.After post-process smoothing,the accumulative error of 60 s reduced from the maximized 20 m to 0.5m.5)Based on the principle of mobile survey,the technology of visual point cloud map and road lane map generation is studied.The local coordinates of the landmark points are obtained by multi-frame image forward intersection,and the local coordinates of the road lane are acquired through homography transformation.Then,the camera pose solved from GNSS/SINS is converted to the visual point cloud and road lane coordinates expressed in ECEF frame.The quality index of visual point cloud is proposed,and the quality of point cloud is improved by data cleaning.It is indicated by error analysis from many aspects that the absolute position accuracy of the road lane is about 10~20cm.6)With the support of high-definition maps with two layers of visual point cloud and road lane,the method of visual positioning and GNSS/SINS/Visual/lane constraint/odometer multi-source information fusion is lucubrated.The mathematical model of different information combination is constructed,in the meantime a visual positioning framework for octomap and DBOW acceleration is proposed.The KITTI dataset test shows that the accuracy of visual positioning yields about 1.5cm and 0.06 deg,and the positioning success rate is close to 100%.The average time cost is only 0.316 seconds,which can meet the real-time requirements.When integrated with INS,a positioning accuracy of 10 cm can be maintained with only one road sign matched successfully in the case of GNSS long time(20min)unlock.Finally,the real-time positioning of GNSS/SINS/lane constraint/mileage fusion in the case of two satellites was tested.For partial block with a duration of 300 s,the positioning accuracy of all of the three components is better than 10 cm.7)A set of high-precision GNSS/SINS data fusion process software named POSMind is independently developed.The software has a rich visual interface,which can not only process GNSS multi-constellation data,but also the data combined with INS,supporting a variety of hybrid process mode inclusive of precise single point positioning(PPP),differential GNSS positioning(DGNSS),loose coupled integration(LCI),tight coupled integration(TCI),semi-tight combination(STC).Furthermore,the forward/backward filters,forward/backward RTS smoothers and combiners is provided to achieve optimal fusion of multiple information.By far,the software is the only one that supports IARPPP/SINS combination function.On this basis,the real-time position and attitude determination algorithm of GNSS/SINS/visual/odometer multi-sensor fusion supported by visual map is realized. |
PDF Full Text Request