Research On Lane Aided GNSS/INS/Odometer Fusion Positioning Technology

With the development of artificial intelligence,unmanned systems such as unmanned vehicles,mobile robots,and unmanned aircraft have become current research hotspots.As one of the important unit of unmanned systems,accurate and continuous positioning is the key to autonomous movement of mobile carriers.For outdoor wheeled-robot carriers,GNSS/INS integrated navigation can meet the positioning needs in most scenarios.However,GNSS positioning accuracy and continuity decline in the environment where the GNSS signal serious deteriorates due to high buildings,trees and other obstruction,which make the inertial navigation errors continue to accumulate and GNSS/INS integrated navigation cannot meet the needs of robots.Using the speed information provided by the odometer and Non-Holonomic Constraints,the degree of error divergence of the integrated navigation system when the GNSS signal is interrupted can be effectively suppressed.However,the accuracy of the positioning cannot be guaranteed under prolonged GNSS outages.Therefore,with the need of precise,continuous and reliable positioning for wheeled-robots in complex environments under deteriorating GNSS signals,this thesis uses the lane information detected by the image from camera as a positioning source.A visual lane aided GNSS/INS/odometer multi-source fusion positioning scheme is studied to achieve lane-level positioning service in an environment with lane.As common markers on the road,the lane lines are uniform and highly distinguishable from the ground,which is a good source of matching location.This thesis studies lane aided multi-source fusion positioning technology,designs a lane detection algorithm assisted by multiple sensor information,studies the lane matching scheme and the two fusion method of lane information in fusion positioning.Besides,the positioning performance was tested and analyzed through experiments.The main research work and contributions of this paper are as follows:(1)For the lane detection in complex environments with shadow,mottled and occlusion,a lane detection algorithm assisted by multiple sensor information is designed.From the image processing,this thesis uses lane edge features to extract lane lines from the image,and design a detection method for the endpoint of the dotted lanes.In order to improve the performance of the detection algorithm,the pose information provided by the IMU and the odometer is used for precisely predict lane.Including two aspects of assistance.Firstly,aided information helps to determine the region of interest of the image combined with the predicted lane.Secondly,aided information helps for lane detection fault detecting based on the consistency between lane,which can reduce wrong detections.Test results in different scenarios show that the lane detection algorithm has high detection accuracy,and can achieve reliable detection precision even in harsh environments,which can meet the needs of fusion positioning system.(2)On the basis of GNSS/INS/odometer fusion positioning algorithm,the lane observation information is introduced to design a lane distance aided multi-source fusion positioning algorithm.First,combined with the self-made lane map,the map matching method of lane is studied.The calculation method of the relative distance between the carrier and the lane lines from the detected pixel coordinates of lane lines is deduced.Based on this,the fusion positioning system is aided by the relative distance between the carrier and the lane lines in the defined lane coordinate system.A lane distance measurement model is established.On the basis of the horizontal position constraint of the solid lanes,the observation information of the endpoints of the lane is added to constrain the position of the carrier in the forward direction.The results of GNSS outage test using open scene data show that the solid lanes can keep the transverse position accuracy of the carrier within 0.35 m,and the statistical error is less than 0.2m(RMS).With the aid of the solid and dotted lanes,the horizontal position accuracy of the system is kept within 0.25 m,and the statistical error is less than0.1m(RMS).Besides,the assist of the lane endpoint can make the positioning less dependent on the odometer.(3)According to the different coordinate system where the measurement is established,a lane pixel coordinate difference aided fusion positioning algorithm under the pixel coordinate system is designed.This algorithm uses the pixel coordinate difference between the projected lane lines and the detected lane lines as the observation of the Kalman filter,and the measurement model is established.The projection and information fusion method of the solid lanes and endpoint of the dotted lanes are respectively designed.The results of GNSS outage test using open scene data show that the lane pixel difference aided positioning algorithm in pixel coordinate system under only the solid lane observation,the transverse position accuracy of the carrier is kept within 0.3m,and the statistical error is less than 0.1m(RMS).With the aid of the solid and dotted lanes,the horizontal position accuracy of the system is kept within 0.2m,and the statistical error is less than 0.1m(RMS).The positioning accuracy of this method is basically equivalent to the lane distance assist method.Besides,test results in a complex environment show that the two lane aided methods can effectively reduce dependence of the positioning system on GNSS positioning quality,and can achieve lane-level positioning accuracy in a long-term deterioration of GNSS.In summary,this thesis carried out in-depth research on the use of lane line as an aided positioning source and the accuracy improvement effect of the positioning system.Lane aided GNSS/INS/Odometer fusion positioning scheme is designed and implemented by this thesis.This scheme can effectively improve the problem that the accuracy gradually declines and deteriorates over time of the fusion positioning system in a complex environment with deteriorating GNSS,and achieves lane-level positioning accuracy for wheeled-robot carrier.It provides algorithm and practice reference for the navigation and positioning scheme of outdoor wheeled-robots.