Research On Positioning Of Driverless Car Based On VISLAM In Low-light Environment

Driverless car must obtain high-precision positioning results before performing tasks such as path planning.At present,driverless car mostly use GNSS to achieve positioning.GNSS has accurate positioning results and does not has cumulative error in an environment without obstructions,but its positioning accuracy is poor in an environment with obstructions.VISLAM can obtain extremely accurate positioning results in environment with sufficient brightness.However,in a low-light environment,there is a problem of reduced positioning accuracy or even inability to locate.This paper adds GNSS positioning information to the VISLAM and proposed the LVG＿SLAM to enhance the positioning accuracy and robustness of driverless car positioning system in a low-light environment.The research contents of this paper are as follows:(1)The driverless car positioning system model based on VISLAM was established: a mathematical model is established for the monocular vision camera;the error model and motion model of the IMU are analyzed;the relevant coordinate system of the self-driving car positioning system was defined and established its conversion relationship;Established the VISLAM back-end optimization model.(2)To solve the problem that VISLAM cannot extract enough feature points in a low-light environment,an RFAST algorithm based on Retinex theory is proposed: use wavelet transform to decompose the image into low-frequency and high-frequency components;Perform noise suppression processing on the high-frequency components of the image;Use image lowfrequency components to estimate the illuminated image,Use the Retinex algorithm on the illuminated image and the denoised image to obtain the enhanced image.(3)Aiming at the low-light environment that is common in the driving process of the driverless driving car,the LVG＿SLAM positioning system is proposed: Based on the VINSMono positioning system framework,a image enhancement module is added to enhance the positioning accuracy and robustness of the VINS-Mono positioning system in a low-light environment.In view of the cumulative error of VISLAM,integrate GNSS information into the positioning system.(4)The EuRoC dataset is used to verify the RFAST low-light image enhancement algorithm proposed in this paper.The Kitti dataset is used to verify the SLAM and GNSS fusion method proposed in this paper.Further verify the effectiveness of the LVG＿SLAM positioning system proposed in this paper,this paper conducts comparison experiments with real vehicles in the real environment for common low-visibility scenes during driving of driverless cars,such as underground parking lot scenes and night road scenes.The results illustrate that the LVG＿SLAM has better performance in positioning accuracy and robustness compared with the VISLAM in a low-light environment.