The Research About Methods Of INS/Stereo-Vision Integrated Navigation Based On Multiple View Geometry

High-precision autonomous navigation technology is one of the key technologies to develop unmanned system platform.Inertial navigation system(INS)becomes the core of the autonomous navigation system as it can provide rich and autonomous navigation information.However,there are inherent weaknesses in inertial navigation systems because the errors accumulate over time.INS alone can not meet the needs of long time autonomous navigation tasks.Due to the rich information of the visual images and the complementary characteristics of visual and inertial sensors,the visual images become a good aiding source of INS.INS / Vision integrated navigation system is the current hot and important development direction in the field of navigation.Aiming at the problem of autonomous navigation of unmanned platform without Global Navigation Satellite System,this paper focuses on the research of stereo vision assisted inertial navigation method based on multi-view geometry.The main work and innovation are as follows:(1)Most of the current inertial/visual navigation systems rely on point features and need large computation and other issues.In this paper,a new method based on multi view geometry constraint is proposed.A unified point and line features observation equation is established by using trifocal tensor,using the image information more reasonable.At the same time,in the process of system operation,the method keeps the system state dimension unchanged,which can avoid the problem of large complexity such as SLAM,multi state constraint Kalman filter(MSCKF)and other methods.The outdoor and indoor experiments show that: a.In the case when the point features are rich and uniform distributed,the introduction of the line feature information make little sense on the overall system accuracy.b.When the point features are not rich enough and line features are rich in the environment,the line observations can greatly enhance the precision of integrated navigation.(2)Based on the characteristics of indoor structured environment,an inertial / stereo vision integrated navigation method based on vanishing point aiding is proposed.At the same time,an improved vanishing points detection method is proposed exploiting the inertial navigation information.The improved method can further improve the detection speed and reduce the error rate.Experimental results show that the proposed method can effectively limit the accumulation of the heading error,and improve the precision of integrated navigation system.(3)A polarized light compass aided inertial / stereo odometry integrated navigation algorithm is proposed.The polarized light compass and stereo visual odometry measurement model and the corresponding linear measurement model are derived;the random cloning extended Kalman filter(SC-EKF)is designed and implemented.The results of outdoor vehicle experiments show that the introduction of polarized light compass information can greatly limit the long and long distance error accumulation.(4)In order to solve the problem that the current calibration methods of inertial/visual integrated systems rely on the calibration target and the operation is complex,a new calibration method of inertial / stereo vision system is presented.The method is based on stereoscopic simultaneous positioning and map construction(SLAM)technology and multiple view batch optimization technology,which can give high precision calibration results without reference to calibration reference.The validity of the algorithm is verified by using the public dataset.