基于多任务辅助学习特征的瞳孔中心检测  

Pupil Center Detection Inspired by Multi-task Auxiliary Learning Characteristic

作　　者：贾静 付高波 赵歆波[1,2] 邹晓春 张宝尚[2] Jia Jing;Fu Gaobo;Zhao Xinbo;Zou Xiaochun;Zhao Baoshang(Northwestern Polytechnical University,Xi’an 710072,China;Science and Technology on Electro-optic Control Laboratory,Luoyan 471009,China)

机构地区：[1]西北工业大学,陕西西安710072 [2]光电控制技术重点实验室,河南洛阳471009

出　　处：《航空科学技术》2023年第9期121-126,共6页Aeronautical Science & Technology

基　　金：国家自然科学基金(61871326);航空科学基金(20185453031,20200051053003)。

摘　　要：眼动跟踪技术在航空领域极具潜力,其作为关键的信息获取和人机交互手段,可用于精确的目标瞄准、飞行员状态监测,对飞行安全和作战精准意义重大。而瞳孔中心检测作为核心技术,决定了眼动追踪系统的鲁棒性和准确性,本文提出一种基于多任务辅助学习特征的瞳孔中心检测算法,模拟人类视觉系统的多任务辅助学习特性,引入多任务模块,实现从粗到精的瞳孔中心检测,并通过试验验证了本文方法的有效性和先进性,从而提高了眼动跟踪的精度和准确性,为航空领域的各项任务和操作提供了更精确、高效和安全的手段。Eye tracking technology has great potential in the aviation field.As a key means of information acquisition and human-computer interaction,it can be used for accurate target aiming and pilot condition monitoring,which is of great significance to flight safety and combat accuracy.As the core technology,pupil center detection determines the robustness and accuracy of the eye tracking system.This paper proposes a pupil center detection algorithm based on multi-task assisted learning features,simulates the multi-task assisted learning characteristics of the human visual system,and introduces a multi-task module to achieve pupil center detection from coarse to fine.Experiments have verified the effectiveness and advanced nature of the proposed method.In turn,it improves the precision and accuracy of eye tracking,providing a more accurate,efficient and safe means for various tasks and operations in the aviation field.

关 键 词：瞳孔中心检测 深度学习 辅助特征 多任务学习 多尺度分类 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]