基于机器学习的视线追踪方法及其在移动端四旋翼控制的应用  

Machine Learning-Based Gaze-Tracking and Its Application in Quadrotor Control on Mobile Device

作　　者：胡佳辉 陆永华[1] 刘江伟 严长凯 刘韬 HU Jiahui;LU Yonghua;LIU Jiangwei;YAN Changkai;LIU Tao(College of Mechanical and Electrical Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing 210016,P.R.China;China Aeronautical Control System Research Institute,Wuxi 214000,P.R.China)

机构地区：[1]南京航空航天大学机电学院,中国南京210016 [2]中国航发控制系统研究所,中国无锡214000

出　　处：《Transactions of Nanjing University of Aeronautics and Astronautics》2023年第5期547-554,共8页南京航空航天大学学报（英文版）

基　　金：supported by the National Natural Science Foundation of China(No.51975293);the Aeronautical Science Foundation of China(No.2019ZD052010);the Zhangjiagang Pre-research Fund of China(No.ZKCXY2101)。

摘　　要：提出了一种基于机器学习的移动设备单目视线追踪技术。这种非侵入、便捷、低成本的注视追踪方法可以实时捕捉用户在移动设备屏幕上的注视点。结合四旋翼飞行器的三维运动控制,将用户的视线信号转化为四旋翼飞行器的控制信号,解决了以往控制方法的局限性,使用户可以通过视觉交互操控飞行器。为了明确该方法的可行性,本文设置了一条复杂的四旋翼飞行赛道。受试者被要求将视线介入到控制流中,以完成飞行任务。通过与基于操纵杆的控制方法进行比较,对飞行性能进行了评估。实验结果表明,本文方法能提高四旋翼飞行器运动轨迹的平滑性和合理性,并能为四旋翼飞行器控制引入多样性、便利性和直观性。A machine learning-based monocular gaze-tracking technology for mobile devices is proposed.This noninvasive,convenient,and low-cost gaze-tracking method can capture the gaze points of users on the screen of mobile devices in real time.Combined with the quadrotor’s 3D motion control,the user’s gaze information is converted into the quadrotor’s control signal,solving the limitations of previous control methods,which allows the user to manipulate the quadrotor through visual interaction.A complex quadrotor track is set up to test the feasibility of this method.Subjects are asked to intervene their gaze into the control flow to complete the flight tasks.Flight performance is evaluated by comparing with the joystick-based control method.Experimental results show that the proposed method can improve the smoothness and rationality of the quadrotor motion trajectory,and can introduce diversity,convenience,and intuitiveness to the quadrotor control.

关 键 词：视线追踪 无人机控制 机器学习 人机交互 眼球驱动 

分 类 号：TP249[自动化与计算机技术—检测技术与自动化装置]