Research On The Fusion Of LiDAR+VIO+GNSS For Precise Positioning In Urban Complex Environment And Verification Of Vehicle-borne Experiment

The Global Navigation Satellite System(GNSS)has entered a new era of multisystem compatibility and common development.The development of multi-frequency and multi-system GNSS has significantly improved the accuracy,reliability and usability of the positioning compared to a single satellite navigation system.It can provide high-precision location services for wisdom logistics,automatic driving and many other intelligent mobile carriers.However,the natural fragility of GNSS makes its positioning performance degraded or even unusable in some complex environments,such as complex blocks,urban canyons,viaducts,and tunnels.But in these GNSS degraded areas,high-precision autonomous navigation and positioning can be achieved by using some relative positioning sensors such as vision,inertial,and Li DAR.Therefore,how to effectively combine the GNSS and the local relative positioning technique to achieve continuous,reliable and high-precision positioning in different environments has important research significance and value.Aiming at the possible degradation or unavailability of GNSS in the urban complex environment,this paper studies the GNSS/Visual/Inertial/Li DAR multi-sensor fusion positioning algorithm to improve the positioning performance of mobile carriers in complex environments.The main work and contributions of this paper are as follows:Firstly,in order to improve the local positioning performance of the multi-sensor fusion algorithm,a graph-optimization based and tightly coupled stereo vision/inertial/Li DAR SLAM is presented in this paper.The pre-processing methods of multi-sensor data,the construction of observation models and the residual equations of different types of sensors,and the joint nonlinear optimization model of multi-sensor data based on sliding windows are studied respectively.On this basis,the complementary characteristics of vision and Li DAR were fully explored and a visionenhanced Li DAR closed-loop optimization algorithm was designed.The accuracy of loop detection was verified in large-scale outdoor scenes.Secondly,facing the GNSS-challenged environment,we further studied the graphoptimization based GNSS/vision/inertial/Li DAR multi-sensor fusion algorithm for precise positioning.The stereo vision/inertial/Li DAR odometry(S-VILO)is used to provide high-precision local relative constraints in global fusion,and the GNSS is used to provide high-precision absolute position constraints.In this way,we achieved a decoupled multi-sensor fusion positioning algorithm.In view of the complex urban environment,the fusion strategy of multi-sensor fusion positioning and the available methods for gross error elimination are discussed.Finally,in order to verify the proposed algorithm in this paper,two groups of vehicle-borne experiment were carried out in urban complex environment.The test results show that the proposed S-VILO algorithm achieves higher pose estimation accuracy compared with the tightly coupled stereo visual-inertial odometer(S-VIO),among which the horizontal accuracy and heading angle accuracy are improved significantly.In large-scale scenarios,the loop closure algorithm proposed in this paper can effectively detect the closed-loop in the trajectory and achieve high-accuracy pose estimation.Moreover,the point cloud map obtained by pose-graph projection has good resolution and global consistency.In terms of multi-sensor fusion positioning,we analyze the PPP/S-VILO and the RTK/S-VILO respectively.The statistical results show that the proposed multi-source fusion positioning algorithm can achieve continuous and smooth positioning results under the condition of GNSS degradation or non-availability,and has good robustness.