Research On Monocular Vision-Inertial Odometry Algorithm Based On Points And Lines Features

The rapid development of SLAM(Simultaneous Localization and Mapping)technology provides good support for robots in autonomous navigation and scene reconstruction and has become a research hotspot in the field of robot navigation.However,it is difficult for visual SLAM to be stable and efficient in fast motion and dynamic scenes and it is difficult to extract the feature information in the low-texture image.Considering the Inertial Measurement Unit(IMU)is complementary to the camera in performance and the line features existing in the environment,this paper uses the visual-inertial based on the point line feature.The Visual Inertial Odometry(VIO)is the research object,and the visual inertia bimodal information fusion research is carried out.The main contents of this article include:Firstly,in view of the contradiction between accuracy and efficiency in the process of feature construction of point and line,image feature extraction and matching are studied.The ORB(Oriented FAST and Rotated BRIEF)corner points and the LSD(Line Segment Detector)line segment are used to extract the point line features and calculate their descriptors respectively to ensure The correct rate of data association.The mechanism of the re-projection error is analyzed and the analytical form of jacobian matrix about camera pose and feature position is derived.Secondly,a VIO initialization scheme based on constant velocity model is proposed to solve the problem of high dependence on initial value of the system during VIO startup,using the frame for calculating camera pose transformation between the gyroscope bias was isolated,and then through the integral observation model respectively to estimate the accelerometer bias,gravity vector,monocular camera scale factor and speed,simplify the time-consuming serious feature extraction and matching process,the experimental results show that the precision initialization method can be in do not break under the premise of raising the speed of the system initialization,guarantee the smooth VIO started.Thirdly,aiming at the problem of dimension explosion and error accumulation in the optimization process,a state variable construction method based on tightly coupled framework and an optimization model based on sliding window are proposed.The state variables include pose,velocity,IMU offset and points line features.The position in the three-dimensional space,the sliding window model only processes the quantitative input data to ensure the solution efficiency of the system.Two optimization strategies are designed according to the current input image type,and the converged state quantity is marginalized into the three-dimensional space,raise the utilization ratio of information.Finally,aiming at the information redundancy problem of VIO system and considering the effect of scene reconstruction,an efficient keyframe screening mechanism is designed and the RANSAC method is used to process the outer point,which reduces the impact of erroneous observations on the overall accuracy of VIO.A sparse composition method combining point line features is proposed,which can better restore the scene structure and texture information compared to the sparse point map.The simulation results show that the above method is superior to the current mainstream algorithm in navigation and positioning accuracy.The actual measurement in the substation environment shows that the VIO system designed in this paper has good practicability.