Research On Self-driving Cars Location Method Based On Fusion Of GPS And VISLAM

As a fundamental capability of self-driving cars,path planning and other tasks highly relies on the accurate localization of the self-driving cars.At present,self-driving cars mostly use GPS to obtain the location of the self-driving cars.GPS can obtain accurate location results in open areas.However,in the environment covered by obstacles,there is a problem that the location accuracy is low or even difficult to locate.VISLAM has become a foundation of indoor mobile robot.However,although the scale and location accuracy have improved markedly in that time,it still can`t handle large-scale complex environment.Therefore,in this paper we propose that incorporation of GPS signals into VISLAM system which can increased robustness,scalability and improved accuracy of localization for the self-driving cars in the outdoor and indoor large-scale environment.The main work as follows:1)Propose an improved algorithm VINS-FAST based on VINS-Mono.Shi Tomasi corner detection for visual pose estimation in VINS-Mono system is not robust enough,this paper proposes to replace the Shi Tomasi corner detection in VINS-Mono visual pose estimation with a more robust FAST corner detection,which can improve system location accuracy.It also proposes a uniform FAST corner distribution strategy to improve the problem of FAST corner detection focusing on areas with obvious features,so that FAST corners can also be detected in areas with weak textures;2)This paper further proposes a location system VINS-GPS based on the fusion of GPS and VINS-FAST.There is a problem that VISLAM’s location accuracy is reduced due to system cumulative errors in a large-scale environment.So we further propose a location system VINSGPS based on the fusion of GPS and VINS-FAST.Using the location results of GPS and VINSFAST system,a nonlinear optimization objective function is constructed,and the location of self-driving cars is solved by nonlinear optimization method;The paper uses the data set Euroc to verify that the FAST corner uniform distribution strategy proposed in this paper can improve the phenomenon that FAST corner detection is too concentrated,In the indoor environment VINS-FAST shows higher positioning accuracy than VINS-Mono.Then on the automated driving data set Kitti,it was verified that the VINS-GPS proposed in this paper can improve the location ability of self-driving cars in outdoor largescale environments.The results show that the algorithm proposed in this paper can not only improve the performance of VISLAM in large-scale environment and it can also improve the location accuracy of GPS in a weak or no signal environment.