Research On Multi-sensor Fusion Positioning Algorithm For BEV Lane Information Coordinatio

Autopilot positioning technology is a hot issue in industrial intelligence research.In complex scenarios such as occlusion,the GNSS(Global Navigation Satellite Syste m)signal may experience accuracy degradation or even loss of lock,making it difficul t to meet the positioning accuracy requirements.To tackle the aforementioned issues,this paper presents a localization algorithm t hat combines multiple sensors and incorporates visual lane information from a Bird’s E ye View perspective.The proposed approach employs the Error State Kalman Filter(E SKF)as the fusion framework and integrates Strap-down Inertial Navigation System(SINS)data and vision-based map matching information to improve the accuracy of th e multi-sensor positioning system.It achieves high accuracy positioning of vehicles in complex scenarios and shows good robustness to meet the requirements of lane level positioning.The main work accomplished in this paper is as follows.(1)The PVQ nominal state propagation of the Strap-down Inertial navigation syst em is derived for the GNSS signal loss lock situation,and the a priori position informa tion of the High Definition Map(HDmap)is fused to achieve a lane level positioning effect with position error at the sub-meter level and attitude error less than 1~0range.(2)For the error correction process of ESKF,according to the data correlation pri nciple of vision-map,the a priori information of HDmap is solved by using BEV lane line detection as loosely coupled observation information,which reduces the system p ositional error and achieves the goal of improving the positioning accuracy.At the sam e time,the overall robustness,flexibility and scalability of the system are improved.(3)Aiming at the problems of serious loss of original lane information and high c omplexity of image processing in the detection process of visual lane lines,the BEV la ne line detection algorithm is proposed to unify the coordinate relationship of visual-m ap matching,and the advantages of strong perception,anti-interference and high effici ency are verified by experiments.(4)In the proposed environment,the feasibility and superiority of the multi-senso r fusion localization algorithm with BEV lane information collaboration were verified by corresponding experimental tests on the positional errors brought by different obser vation information to achieve lane level localization according to the lane level naviga tion requirements of the multi-sensor fusion localization algorithm.In view of the above work,this paper finally completes the design of multi-senso r fusion positioning system for BEV lane information collaboration,and realizes the ac tual verification of the algorithm in the proposed complex environment,which meets t he requirements of continuous,reliable and high-precision lane-level positioning.