Unmanned aerial vehicle takeoff/landing guiding-oriented deep learning label data generation method

Abstract

The invention discloses an unmanned aerial vehicle takeoff / landing guiding-oriented deep learning label data generation method. According to the method, labeling requirements and distributing tasks are defined; users log in labeling clients; the labeling tasks and labeling requirements are received through the labeling clients; each to-be-labelled scene image is manually labelled; the labelled scene image are stored to a database system via an xml format; and the database system is updated in real time. After all the to-be-labelled scene images are completely labeled, an assessor logs in an assessor client and accesses the database system through an assessor client network, and the assessor client assesses the labeling results (namely, each labelled scene image). According to the method, labeling tasks are issued through a network manner, an assessment method is automatically designed to assess the labeling result, so that the data labeling efficiency and labeling result reliability are greatly improved, and the reality demands for deep learning large-scale sample labeling are effectively solved.

Description

technical field [0001] The present invention mainly relates to the design field of a UAV autonomous take-off and landing guidance system, in particular to a deep learning label data generation method for UAV take-off and landing guidance. Background technique [0002] The UAV take-off and landing guidance system is designed to solve the problem of autonomous take-off and landing in weak or GPS-denied environments. The guidance system obtains the scene image containing the UAV target during the take-off and landing process of the UAV through the camera, and extracts the UAV target area and anchor point coordinates in the image, and uses computer vision measurement and filter estimation to solve the UAV's position. The world coordinate pose, so as to realize the autonomous take-off and landing of the guided UAV. Extracting the UAV target area and anchor point coordinates from the image is a necessary function of the guidance system. [0003] Aiming at the shortcomings of the...

AI Technical Summary

Problems solved by technology

[0003] Aiming at the shortcomings of the method of extracting UAV target area and anchor point coordinates from features such as corners and edges, there are shortcomings such as weak applicability and sensitive parameters, and a deep learning scheme is proposed to remove parameter dependence and improve scene applicability

Images