Research On Robot SLAM Based On Vision And Ins Fusion In Indoor Environment

In the intelligent information today,robotics has gradually matured.Some service robots have slowly begun to be accepted by the general public and entering people's work and life.Most functions of service robots are based on autonomous navigation,path planning and obstacle avoidance technologies.The premise of function realization is that the robot simultaneous localization and mapping in an unknown environment,that is called SLAM(simultaneous localization and mapping).However,when using vision sensors alone for simultaneous localization and mapping,accurate vision data often cannot be obtained in indoor environments with less characteristic textures such as white walls and glass curtain walls,which makes it difficult to obtain accurate positioning and maps.In order to improve the simultaneous localization and mapping accuracy of the robot in an indoor environment,this thesis uses a binocular camera and low-precision inertial navigation as sensors,and studies a robot simultaneous localization and mapping algorithm that fuses the two sensor information,the main contents are as follows:In the visual information processing part,first,by modeling the binocular camera,the conversion relationship between the two-dimensional coordinates of the image and the actual three-dimensional coordinates is derived.Then,ORB(oriented fast and rotated brief)feature points are extracted from the visual image,and the feature points between the binocular cameras are brute-force matched,and the feature points between the frame and frame image are quickly matched by FLANN(fast approximate nearest neighbor search library).Then use the random sampling consistency algorithm to remove the erroneous matching feature points.Finally,the image matching experiment shows that all feature points can be matched correctly.Based on the multi-state constrained Kalman filter algorithm,the inertial navigation pose integral value is used as the camera pose estimation value,and multiple camera historical poses are added to the sliding window as the system state,and the visual reprojection error is used as the system observation model,Filter and update the system state,and finally get accurate camera pose and map,so as to realize the realtime SLAM of the robot.In order to solve the problem that the whole system cannot run when a single inertial navigation fails,a multiple inertial navigation redundancy algorithm is designed.Based on the extended Kalman filter algorithm,the system state of the multi-inertial navigation system is established,and the pose residuals between the inertial navigation systems is used as the observation model to filter and update the system state.A residual fault detection function is established to detect whether the inertial navigation is faulty.When the main inertial navigation fails,the auxiliary inertial navigation can quickly replace the faulty main inertial navigation,so that the system can continue to operate.Through the KITTI(karlsruhe instiute of technology and toyota institute)data set test and building an experimental platform to test in the actual environment,the SLAM algorithm proposed by the text is tested.The experimental results show that the accuracy of the SLAM algorithm can meet the design requirements of the robot,and the system can continue to run when the inertial navigation fails.