In order to improve the robustness and accuracy of the visual simultaneous localization and mapping(SLAM)system,a visual SLAM algorithm based on mismatch rejection and ground constraints is proposed.Firstly,the feature point pair rejection mechanism is introduced in the front-end of SLAM,and the matching quality is analyzed by counting the distances between feature point pairs and the distribution characteristics,and then the mismatched feature points are eliminated.Secondly,the RGBD camera is used to extract ground vectors,and the ground vector constraints are introduced into the back-end optimization process of SLAM,which can effectively prevent over-optimization of the z-axis and suppress the drifting of z-axis,and improve the accuracy of position optimization.Finally,the experiment and analysis are carried out on the public datasets.The experimental results show that compared with the ORB-SLAM2 algorithm,the absolute trajectory error of the mismatch rejection algorithm is reduced by 17.63%on average,and the absolute trajectory error of the ground constraint algorithm is reduced by 39.20%on average,which verifies that the proposed algorithm has better accuracy and robustness.
machine visionsimultaneous localization and mappingfeature point rejectionback-end optimizationground constraints
NETL
NSTL
万方数据
