Autonomous Scene Matching Navigation Based on Range Optimization of Reference Image by Interframe Matching

Abstract

The traditional scene matching assisted navigation is closely coupled with the offline planning route. It is mainly the route guidance mode. When the mission is changed, the interception or the sudden air situation causes the deviation from the surveying and mapping channel, the scene matching and positioning will fail. It does not have pure visual navigation and positioning ability. This paper focuses on an autonomous scene matching navigation method that can be decoupled from offline planning routes. Combined with the control idea of continuous flight system, we preliminarily estimate the navigation information through the sequence real-time inter-frame registration, and optimize and evaluate the satellite remote sensing reference map combined with the current estimated pose information, and scale and crop the reference map area to be registered. Then, the real-time map at the current moment is matched with the optimized high-similarity area reference map, which can effectively reduce the amount of matching calculation, improve the matching stability, and realize autonomous scene matching navigation and positioning that is decoupled from the planned route. After the system software and hardware design, based on the multi-rotor UAV and non-real-time Google satellite map, a free flight navigation test was carried out in Chang'an Park of Xi'an. When the flight height is 500m above the ground, the optimized reference map area and the real-time map overlap >90%, the average horizontal positioning error of pure visual scene matching navigation is 16.4m (<4% of the height), and the average altitude positioning error is 26.9 m (<6% of height), the average yaw angle error is 2.8° (<5°), and the update frequency is 20Hz, which can meet the technical requirements of high cohesion and low coupling for autonomous navigation of UAVs in GPS-denied environments.