UAV Application And Research For VIO Autonomous Localization

Simultaneous localization and mapping(SLAM)technology is very critical for robot localization and navigation.Vision sensors can’t overcome scenes with insufficient illumination,missing texture loss,and rapid motion.Rapid response is the characteristic of the Inertial measurement unit(IMU),and IMU is characterized by rapid response and not affected by image quality.The position and attitude is calculated through the principle of inertial navigation.Therefore,the complementary vision sensors for localization of the visual sensor and IMU form a more robust SLAM solution,which is called Visual-Inertial Odometry(VIO).It is widely used in weak GNSS(Global Navigation Satelite System)Signal scene.Through this technology,the intelligent robot accurately locates itself and performs related tasks.This thesis proposes a tightly-coupled binocular visual inertial odometry based on nonlinear optimization using binocular camera images and IMU data,which is acquired by RGBD sensors for autonomous localization under conditions of dim light,dynamic blur,and fast motion.The tightly coupled VIO system is deployed to the UAV,and in order to closely use the available sensor data,the extended Kalman filter technology is used to fuse the IMU data of the UAV into the system.By Loosely coupling the highfrequency and high-precision IMU data of the UAV itself with the VIO system,we build an aerial vehicle hardware and software platform with both tightly and loosely coupled technologies.The main research contents include:Firstly,performes joint calibration of the camera-IMU of the RGBD sensor,and align the time stamp and coordinate system of the binocular infrared camera and IMU.the initialization and back-end optimization of VIO is analysed to get the optimized position,speed and direction for the back-end by constructing an optimization function Optimize by Bundle Adjustment.Secondly,the VIO system was transplanted to the aircraft through hardware selection and software construction.It compensates the slightly lower frequency and low accuracy of the IMU carried by the RGBD sensor by adding the aircraft’s own IMU data and using the extended Kalman filter technology for loose coupling,which increasing the pose output from 15 Hz to 50 Hz.Thirdly,to detect the target during initialization,we choose the traditional target detection based on a priori information and the filtering-based KCF algorithm.The tracking algorithm was used to track it.An application system for autonomous positioning of drones based on multi-sensor fusion was constructed to visualize the track path through interpolation.The binocular VIO algorithm improves the accuracy by more than 20%,compared to the current mainstream ROVIO and VINS-MONO algorithms,under the data set by comparing the absolute positional error-related indicators and the global consistency of the flight trajectory.In the room,the flight pose is obtained by fusing the UAV’s own IMU data,and it obtained by the indoor visual capture device compared with the true value.the effect was improved by the binocular VIO algorithm.And it can achieve indoor target detection,tracking,and waypoint generation applications in the weak GNSS conditions.