摘要
眼球具有高速、高精度的运动能力,不同的眼球运动模式会导致视线轨迹和注视位置的变化,这会改变眼球相对于测量仪器的位置和角度,在面对迅速变化或遮挡等特殊情况时,传统方法缺乏对眼动规律的全面把握,无法准确测定角膜中心与瞳孔间的真实连接方向,这导致了补偿效果的准确性和稳定性受限。为此,提出基于机器学习的视线跟踪误差自动复合补偿方法。基于眼位测定仪结构,建立眼部视线的方向映射模块;基于眼动规律,测定角膜中心与瞳孔间真实连接方向,在3D空间体系内完成眼位测定仪视线中心跟踪定位。利用三维空间体系建立测量坐标,通过三维视线方向,计算眼位测定仪视线跟踪误差。将误差输入BP神经网络,在机器学习导向下根据各层感知器的反馈输出,建立激活函数计算输出层总误差,通过损失函数实现视觉跟踪误差自动复合补偿。经实验证明,所提方法能有效补偿不同视线移动条件和眨眼情况下的眼位测定仪产生的视线跟踪误差,具备一定的实用性。
The eyeball has the ability to move at high speed and with high precision.Different eye movement patterns can lead to changes in the trajectory and fixation position of the eye,which can change the position and angle of the eyeball relative to the measuring instrument.When facing special situations such as rapid changes or occlusion,traditional methods lack a comprehensive understanding of eye movement patterns and cannot accurately determine the true connection direction between the corneal center and the pupil,This results in limited accuracy and stability of the compensation effect.To this end,a machine learning based automatic composite compensation method for line of sight tracking error is proposed.Based on the structure of the eye position measuring instrument,establish a directional mapping module for the eye line of sight;Based on the law of eye movement,determine the true connection direction between the corneal center and the pupil,and complete the tracking and positioning of the eye position measurement instrument's line of sight center in a 3D spatial system.Using a three-dimensional spatial system to establish measurement coordinates,calculate the line of sight tracking error of the eye position measuring instrument based on the three-dimensional line of sight direction.Input the error into the BP neural network,and based on the feedback output of each layer’s perceptron under the guidance of machine learning,establish an activation function to calculate the total error of the output layer,and achieve automatic composite compensation of visual tracking error through the loss function.Experimental results have shown that the proposed method can effectively compensate for the line of sight tracking errors generated by eye position measuring instruments under different conditions of line of sight movement and blinking,and has certain practicality.
作者
吴来新
王志
WU Laixin;WANG Zhi(China-Wenzhou International Innovation Center of Ophthalmology&Optometry Wenzhou,Zhejiang 325024,China)
出处
《自动化与仪器仪表》
2024年第10期370-374,共5页
Automation & Instrumentation
基金
浙江省重点研发计划5G信息技术驱动的人工智能远程诊疗系统研究(2020C03111)。
关键词
机器学习
眼位测定仪
视线跟踪
误差补偿
BP神经网络
machine learning
eye position measuring instrument
eye tracking
error compensation
BP neural network
