| With the development of national economy and modern science and information technology,mobile robots are more and more widely used in our daily life,even in the field of national strategy,which is of great significance to human society.Accurate positioning in unknown environment determines whether the mobile robot can successfully carry out autonomous navigation and complete various tasks.Therefore,based on the research background of binocular inertial vision sensor in complex environment,this paper studies indoor positioning technology based on binocular inertial vision odometry,so as to provide accurate positioning information for robots working in complex environment.The main contents and innovations of this paper are as follows:Firstly,this paper proposes a method of binocular visual odometry based on similar structural constraints and adaptive adjustment strategy.In order to improve the accuracy and speed of feature point matching,this paper improves the traditional RANSAC algorithm by using the similar structural constraints of feature points to improve the matching accuracy and efficiency;In order to keep a good balance between the quantity and quality of the tracked feature points and improve the accuracy of pose estimation,an adaptive adjustment strategy is proposed,which dynamically adjusts the quality control threshold of feature matching by tracking the survival rate of feature points.In order to reduce the possibility of false closed loop in closed-loop detection,a closed-loop detection method based on hybrid similarity is proposed.Finally,the algorithm is simulated with Kitti data set,and compared with the classic ORBSLAM2 algorithm.The results show that the average positioning accuracy is improved by 21%,which meets the requirements of real-time positioning.Secondly,this paper proposes a method of binocular inertial and visual odometry based on improved FLD point line features.In order to improve the average speed of line feature extraction,this paper uses FLD algorithm to replace the traditional LSD algorithm.In view of the defect that FLD algorithm is easy to extract short line features in complex scenes,a line feature constraint elimination criterion is designed to retain the relatively high quality long line features,so as to improve the speed of the extraction algorithm.The average extraction time is62.8% less than LSD.In order to further improve the quality of line feature extraction,this paper combines the line features which are closer to each other to avoid detecting a large number of similar line segments in some edge regions,which can reduce the complexity of line feature matching.Aiming at the problem that the existing visual inertial odometry system based on point feature is difficult to extract and easy to lose point feature in the dark,weak texture and other complex environment,this paper combines the point line feature with IMU pre integration technology,graph optimization and other means to design a more perfect binocular visual inertial odometer calculation method based on point line dual feature.At the same time,the computational complexity of the algorithm is reduced by sliding window optimization,which ensures the accuracy of the algorithm and makes the algorithm run in real time.Finally,the simulation results on euroc dataset and the comparison with several open source algorithms show the effectiveness of the proposed algorithm.Thirdly,the prototype experiment in the real environment is carried out on the platform of multi-sensor fusion,which verifies that the vision inertial integration positioning system designed in this paper can provide more accurate pose information for the mobile robot in the complex and unknown environment.In the experiment,the running results of visual inertial integration system under ideal environment,illumination change and weak texture environment are tested.The experimental results show that the system designed in this paper can provide a relatively stable and high-precision pose information in the above complex environment. |
PDF Full Text Request