ORBSLAM2-based unmanned aerial vehicle autonomous navigation method

Abstract

The invention discloses an ORBSLAM2-based unmanned aerial vehicle autonomous navigation method, and belongs to the field of visual synchronous localization and map construction. The method is mainly composed of visual front-end and visual rear-end global optimization and mapping and three-dimensional path planning. The visual front-end mainly comprises ORB feature point extraction and matching and calculation of a current unmanned aerial vehicle pose by a visual odometer, and the part realizes a positioning function. Visual rear-end global optimization and mapping mainly comprise BA global graph optimization and dense point cloud image mapping, and the part realizes global optimization and mapping functions. Three-dimensional path planning realizes indoor and outdoor obstacle avoidance operation of the unmanned aerial vehicle, and the part realizes a path planning function. ORB pixel points are extracted by using improved FAST angular points, and ORB feature points are subjected to block optimization. Pose estimation is made by using an improved ICP-PnP algorithm.

Description

technical field [0001] The invention relates to an autonomous navigation method for an unmanned aerial vehicle based on ORB_SLAM2, and belongs to the field of visual synchronization positioning and map construction (SLAM). Background technique [0002] Unmanned aerial vehicle has a very wide range of applications in many fields. It is a kind of mobile robot, referred to as UAV (Unmanned Aerial Vehicle, UAV). By carrying a controller on the UAV, people can control the aircraft on the ground. flight. At the same time, the advent of the Simuitaneous Localization and Mapping (SLAM) algorithm is to solve the positioning problem of the robot when there is no GPS signal, and it can also measure the map of the environment. Autonomous navigation of drones relying on RGB-D is mainly divided into three elements: "Where am I?", "What is my environment like?", "How do I go to get to the mission destination?", this The three problems represent respectively: positioning mapping and path ...

AI Technical Summary

Problems solved by technology

[0003] For example, a binocular 3D dense mapping method based on ORB_SLAM2 published in Patent No. CN 108520554 A mainly proposes four ORB (Oriented FAST and Rotated BRIEF) threads for SLAM mapping: tracking thread, local map thread, closed-loop monitoring thread and dense Mapping process, this method puts the ORB feature point extraction and matching process in the tracking process, if the ORB feature points appear in a pile, it will cause a large deviation in the visual odometer to judge the pose, and the ORB feature point is not divided into blocks optimization, it will lead to a large positioning error

[0004] In the existing UAV SLAM system, after extracting ORB feature points in the positioning module, it is necessary to consider the existence of feature points in piles, which affects the estimation of the visual odometry pose. After the dense point cloud image is generated in the mapping module, there is no global optimization. Errors occur when the image is transformed into pixel coordinates. The traditional RRT* algorithm may cause the UAV to hover in the corner and cannot avoid obstacles or the circle around the obstacle is too large when the path planning module is running.

Images