Research On VIO Positioning System Based On Improved Point-line Feature

Visual inertial odometer VIO(Visual Inertial Odometry)is often used to solve the real-time positioning problem of mobile robots.The current mainstream VIO systems usually use feature points or line features as visual constraints.However,traditional line features are based on descriptors for line feature matching and there is a problem that line segments are segmented,resulting in a long time-consuming line feature matching.At the same time,the traditional VIO system has a small number of feature matching in common complex scenes such as fog,resulting in low positioning accuracy and poor robustness of the system.In view of the above problems,this paper proposes an improved point-line feature VIO system: in online feature matching,the method based on feature point constraints and line feature geometric constraints is used to improve the number and speed of line feature matching;in image preprocessing,adding The undercurrent channel combines the automatic color scale dehazing algorithm to improve the number of feature matching of the system in foggy scenes.In the front-end feature matching,feature point matching and line feature matching are combined to improve the positioning accuracy and robustness of the system in the absence of feature points.One is based on a variety of feature point matching efficiency comparison experiments,using the feature point matching method of FAST feature points plus optical flow.The second is to add a line segment merging algorithm to solve the problem of line segment segmentation during online feature extraction.Third,when matching online features,a line feature matching method based on feature point constraints and line feature geometric constraints is proposed.At the same time,a FAST feature point addition algorithm is proposed to ensure that there are enough feature point constraints for line feature matching.Experimental results prove that the matching speed and matching quantity of this line feature matching method are about two times higher than those of the traditional matching method.In order to improve the number of feature matching of the system in foggy scenes,in the visual preprocessing part,a dehazing algorithm combining dark channel and automatic color scale is added.Experimental results show that after dehazing,compared with before dehazing,the number of feature points that can be matched is more than twice.After adding the undercurrent channel combined with the automatic color level defogging algorithm,the feature matching has better robustness.In the back-end optimization,the feature point observation model,the line feature observation model and the IMU observation model were constructed,and the visualinertial tight coupling optimization algorithm was used to optimize the pose estimated by the observation model.Through the loop detection function,the pose estimation was reduced.cumulative error.Finally,this paper builds a complete VIO dataset simulation experiment and actual experiment,and conducts performance tests in the EUROC dataset,weak texture corridor and outdoor loopback dataset and the actual environment,and its positioning accuracy and real-time performance have a certain degree of improvement.promote.