Research On Visual-inertial Odometry Algorithm For UAV

In recent years,with the emergence of emerging products such as unmanned aerial vehicles(UAV),unmanned vehicles,robots,and AR,Simultaneous Localization and Mapping(SLAM)has become a focus of attention and research.Point features are widely used visual features in visual SLAM,but the extraction and tracking of point features is challenging in non-textured or light-changing environments.At this time,line features are a good alternative.Line features are a higher level of visual features than point features,which can provide us with richer geometric information in the environment.However,because of the low frame rate,vision sensors cannot handle dynamic scenes.The Inertial Measurement Unit(IMU),because it has an accelerometer and a gyroscope,can measure its own angular velocity and acceleration,thus complementing the vision sensor.The combination of vision sensor and IMU has formed a more robust SLAM scheme,also known as vision-inertial odometer.In this article,a point-line feature extracted from a monocular camera image and a three-axis accelerometer and gyroscope provided in the IMU are used to propose a visualinertial odometry based on the comprehensive features of point and line.Positioning and pose estimation.Starring research results include:(1)Offline calibration of the camera and IMU,and offline calibration of the cameraIMU to obtain the parameters of the sensor,so as to configure the parameter files required by the system for testing the system in an actual scenario;(2)In the process of extracting online features,the LSD(Line Segment Detector)feature and the LBD(Line Band Descriptor)descriptor are improved to improve the line segment extraction process due to fuzzy noise and other influences to break;(3)The line features adopt the Plueck coordinate form in the epipolar geometric constraint,and the orthogonal form is used in the optimization process to reduce the calculation load of the equipment,and the corresponding form of the measurement residual of the wire feature is derived;(4)A sliding window model with IMU pre-integration constraints and point / line features is used to tightly couple vision and inertial sensor data;(5)In addition to point features and IMU in the system optimization framework,add line features to improve the robustness of the entire system;Finally,experiments and accuracy analysis are performed on the constructed system,which verifies the effectiveness of the proposed system scheme.The relative pose error and absolute pose error of the system are analyzed and compared with the mainstream visual inertia scheme VINS-mono.