Font Size: a A A

Research And Implementation Of Localization Technology Based On Global Semantic Map

Posted on:2021-03-19Degree:MasterType:Thesis
Country:ChinaCandidate:J T WuFull Text:PDF
GTID:2428330620964027Subject:Engineering
Abstract/Summary:PDF Full Text Request
In the field of localization,cameras,GPS,lidar and other sensors are usually used together for localization.The pure visual localization technology only with camera to be the sensors is generally considered to be of no practical value due to its low localization accuracy and poor robustness.However,GPS has the problem of obscured signals.Lidar is expensive and will greatly increase the cost of the localization system.Pure visual localization technology does not have those above problems.Therefore,it is of great practical significance to find a pure visual localization technology with high localization accuracy,good robustness,and real-time performance.In this paper,a pure visual localization technology based on global semantic map is designed and implemented on a heterogeneous embedded platform,which achieves better localization accuracy,robustness and real-time performance.And an optimization algorithm is proposed to reduce the negative impact of dynamic objects,similar objects and environment changes on localization accuracy.The main work and innovation of this paper are as follows:1.A pure visual localization technology based on global semantic map is designed and implemented on a heterogeneous embedded platform,which achieves better localization accuracy,robustness and real-time performance.2.Explore the feasibility of optimizing localization accuracy with nonlinear optimization algorithms,and adopt nonlinear optimization algorithms to further optimize localization accuracy.3.The RANSAC algorithm is used to remove data noise for better localization accuracy.4.An optimization algorithm is proposed to reduce the negative impact of dynamic objects,similar objects and environment changes on localization accuracy,which has been deployed in the localization algorithm.The average localization error of the localization system in this paper is 0.24 m on the KITTI dataset 00 test sequence and 0.13 m on the KITTI dataset 05 test sequence.And the average localization frames rate of the localization system is 10 frames per second.In the end,a low-cost,practical visual localization technology has been realized in this paper.
Keywords/Search Tags:Global semantic map, pure visual localization, heterogeneous embedded platform, localization accuracy optimization
PDF Full Text Request
Related items