Research On Key Technologies Of Visual/Inertial Integrated Navigation For UAV

Due to the diversification and complexity of autonomous navigation requirements and application environments,visual/inertial integrated navigation with good autonomy,high precision and low cost has become a research hotspot.Compared with single visual navigation or inertial navigation,the integrated visual/inertial navigation method can solve the problem of poor navigation effect in special scenes and inertial navigation cumulative error,thus it is nessesary to improve the robustness and accuracy of the system.Based on this background,this thesis studies and improves the algorithm related to the visual/inertial integrated navigation,and designs a visual/inertial integrated navigation system for UAV.Through the experiment of the public data set named EuRoC MAV and the outdoor experiment,the effectiveness of the integrated navigation system is analyzed and verified.Since the camera is prone to image blur when moving,affecting the positioning accuracy of the visual navigation,it is necessary to deblur the blurred image.Aiming at the defects of traditional image deblurring algorithm,such as long time-consuming and high algorithm complexity,an improved image deblurring algorithm based on Retinex algorithm is proposed.Firstly,the original blurred image is edge-enhanced by Retinex algorithm,and then the iterative alternating optimization method is used to perform the estimation of the restored image and the blurred kernel,and finally deconvolution is carried out to obtain the final clear image.Experiments show that the improved deblurring algorithm can reduce the complexity of the algorithm and reduce the algorithm running time.Since the core algorithm of visual odometry is the detection,tracking and matching of image feature points,and the matching accuracy will directly affect the accuracy of carrier motion estimation,it is necessary to improve the efficiency and quality of feature points detection and matching.In this thesis,an improved monocular visual odometry method based on SVD is proposed for the low precision and long time of traditional positioning algorithms.Firstly,the image is compressed by SVD,then the feature points are detected,and then the NCC and RANSAC algorithms are used to match the feature points.and finally estimate the motion of the carrier.The experimental results show that compared with the traditional visual odometry method,the improved algorithm is not only high precision,but also low time-consuming.Due to the large amount of computation caused by repeated propagation of IMU measurements in visual/inertial integrated navigation,the IMU measurement is performed by IMU pre-integration method,and then the tightly coupled monocular visual inertial odometry is designed through the joint initialization of vision and IMU.and finally the global pose graph optimization is completed.In order to verify the performance and practicability of the integrated navigation system,the EuRoC MAV public data set is used to verify the high precision and robustness of the system.Then a PC software for visual inertial data acquisition is designed.The software is used to collect data from outdoor experiments,and the effectiveness of the integrated navigation system is verified by using outdoor experimental data.The final experimental results prove that the integrated navigation system of this thesis not only has high precision but also high robustness.