UAV Vision Fusion Navigation Technology Based On Scene Matching

Unmanned Aerial Vehicle(UAV)navigation system is the key to ensure the safe flight of UAVs,currently the most widely used UAV navigation methods is the combined navigation of Global Navigation Satellite System(GNSS)and Inertial Navigation System(INS).However,INS cannot be used alone because of its cumulative error.Under denied conditions,GNSS cannot provide accurate position navigation information.This thesis focuses on the inherent limitation of GNSS and proposes a vision fusion navigation and positioning method of UAV based on scene matching in low and medium altitude flight environments.The vision-assisted UAV navigation system needs to consider the following three issues:First,because UAV flight is a dynamic and continuous process,the speed of visual position estimation is required to be fast;Second,the accuracy of visual position estimation should be high enough to ensure the accuracy of UAV flight position;Third,in order to ensure the safety of UAV flight process,the stability of visual navigation algorithm is necessary.Based on the above three issues,this thesis investigates vision fusion navigation method of UAV based on scene matching,and carries out the following four aspects of research work:(1)Since the airborne digital image maps are generally very large in size with huge data volume,the efficiency for general matching algorithms is relatively low to meet the demand of navigation and positioning.In order to improve the speed of the algorithm,this thesis proposes a fast image matching method based on the scene a priori feature database,which improves the efficiency of the algorithm while ensuring the matching accuracy.Meanwhile,with the foundation of fast image matching method,this thesis proposes a visual pose estimation method based on scene matching,which provides a feasible solution for the UAV visual navigation strategy.(2)In order to demonstrate the dynamic continuous localization process of UAV visual navigation,two inter-frame processing methods,feature points optical flow tracking and keyframes matching,are proposed in this paper.The inter-frame processing method based on feature points optical flow tracking avoids the feature extraction and matching process and directly obtains the matching points pairs of the given image frame.The inter-frame processing method based on keyframes matching uses matching between keyframes instead of matching between UAV image and digital image map,which reduces the number of matching between UAV image and digital image map and greatly improves the efficiency of visual pose estimation.The two different inter-frame processing methods improve the efficiency of the algorithm while ensuring the accuracy of the algorithm.(3)This thesis proposes a Kalman filter-based "vision+INS+altimeter" multi-source data fusion and cumulative error correction method to ensure the stability of the algorithm.According to the different processing methods among frames,the fusion methods consists of two types of real-time fusion pose estimation methods,one based on feature points optical flow tracking and the other one based on keyframes matching.Experiments under different scenes show that the algorithm meets the requirements of accuracy and speed while ensuring stability.(4)The demonstration software of UAV vision fusion navigation technology based on scene matching is designed to visualize and simulate the proposed "vision+INS+altimeter" multi-source data fusion method.The algorithm is partially verified by conducting experiments using the DJI M100 UAV of our lab.After the simulation and flight experiments,the UAV vision fusion navigation method based on scene matching proposed in this thesis is proved to be able to control the error within a certain range and meet the requirements of UAV vision navigation.