Real time vision-based auxiliary positioning method for UAV airborne processor

UAVs have a wide range of application value in border and coastal defense, river inspections and other fields. However,the inspection process of drones will lead to unreliable positioning information due to satellite signal interference or signal deception, which will seriously affect the credible inspection of drones. In response to the above problems, this paper studies the UAV auxiliary positioning system based on image algorithms, and proposes a fast scene matching algorithm named Onboard match. First, Gaussian pyramid is used to preprocess the image to improve the speed of subsequent feature point extraction, and then combined with DOH features Extraction, BRISK for feature description, LSH for feature matching and PROSAC for removal of mismatches compose the overall matching process. It has been experimentally verified that Onboard match achieves real-time and accurate scene matching on the airborne processor, which can effectively assist the UAV's credible inspection mission. Experiments have verified that the matching speed of Onboard-match algorithm on the TX2 onboard is 1.6 times faster than ORB, the most commonly industry-used real-time matching algorithm, and the average matching accuracy rate is 90.5% higher than ORB. The actual verification result on the airborne processor proves that the matching position error of Onbaord match is 4.42m, and the average frame rate on the airborne processor is 21.4FPS, which can realize real-time and accurate scene matching and realize the auxiliary positioning function of the UAV in the system.