数据与计算发展前沿 ›› 2026, Vol. 8 ›› Issue (5): 202-216.

CSTR: 32002.14.jfdc.CN10-1649/TP.2026.05.016

doi: 10.11871/jfdc.issn.2096-742X.2026.05.016

• 技术与应用 • 上一篇    下一篇

基于EMA与SCConv的特征融合人眼视线估计方法研究

李为1,2,*(),杨磊1,3,黄智豪1,4,赵国成5,赵旭东5   

  1. 1 珠海翔翼航空技术有限公司， 广东 珠海 519000
    2 北京航空航天大学， 航空科学与工程学院， 北京 100044
    3 北京交通大学， 电气工程学院， 北京 100044
    4 西北工业大学， 航空学院， 陕西 西安 710072
    5 江苏普旭科技股份有限公司， 江苏 南京 210000
  • 收稿日期:2026-04-27 出版日期:2026-10-20 发布日期:2026-10-20
  • 通讯作者: *李为（E-mail：li5190001@163.com）
  • 作者简介:李为，珠海翔翼航空技术有限公司，飞行专家, 在读硕士，副高级飞行员，主要研究方向为人机环境工程。
    本文承担工作为：模型设计，算法实现，实验验证。
    LI Wei, employed at Zhuhai Xiangyi Aviation Technology Co., Ltd., is a flight expert, an associate senior pilot, and is currently pursuing a master’s degree. His primary research focuses include human-machine environment engineering.
    In this paper, he is responsible for model design, algorithm implementation, and experimental verification.E-mail：li5190001@163.com
  • 基金资助:
    中国南方航空集团有限公司科技创新项目（综合多模态数据的飞行员认知状态评估系统）(KJ-FA-2025-0003)

Research on Gaze Estimation Method Based on EMA and SCConv Feature Fusion

LI Wei1,2,*(),YANG Lei1,3,HUANG Zhihao1,4,ZHAO Guocheng5,ZHAO Xudong5   

  1. 1 Zhuhai Xiangyi Aviation Technology Co., Ltd, Zhuhai, Guangdong 519000, China
    2 School of Aeronautic Science and Engineering, Beihang University, Beijing 100044, China
    3 School of Electrical Engineering, Beijing Jiaotong University, Beijing 100044, China
    4 School of Aeronautics, Northwestern Polytechnical University, Xi’an, Shanxi 710072, China
    5 Jiangsu Puxu Technology Co., Ltd., Nanjing, Jiangsu 210000, China
  • Received:2026-04-27 Online:2026-10-20 Published:2026-10-20

RichHTML

2

PDF

2

摘要：

【目的】解决无约束环境下因头部大角度偏转与复杂光照干扰导致人眼视线估计精度下降的问题。【应用背景】面向复杂条件下的人机交互、疲劳驾驶监测等应用场景，主要针对自然环境中的眼部图像分析和视觉感知任务。【方法】构建一种融合高效多尺度注意力（EMA）与空间通道重构卷积（SCConv）的双分支网络；通过残差网络和改进的动态姿态感知Agent Swin Transformer分别提取局部和全局特征，利用SCConv抑制冗余，借助EMA增强多尺度表征，并结合头部姿态先验进行跨模态补偿。【结果】在MPII FaceGaze数据集上平均视线角度误差降至4.23°；在极端偏转区域（|Yaw|>15°且|Pitch|>15°）误差仅5.8°；在更具挑战的ETH-XGaze数据集上取得了4.41°的优异表现，展现出强劲的跨数据集泛化能力；此外，模型单帧推理延时仅为16.1 ms，推理帧率高达62.1 FPS。【结论】该特征融合方法有效解决了视线估计外观法中的特征冗余与尺度敏感问题，在兼顾工业级高实时性的同时，显著提升了极端姿态与复杂场景下的估计精度与鲁棒性，具有极高的工程应用价值。

关键词: 视线估计, 特征融合, EMA模块, SCConv模块, 头部姿态, 鲁棒性

Abstract:

[Purpose] This study aims to address the problem of decreased accuracy in human gaze estimation in unconstrained environments caused by large-angle head rotations and complex illumination interference. [Application Context] It targets application scenarios such as human-computer interaction under complex conditions and fatigue driving monitoring, focusing primarily on eye image analysis and visual perception tasks in natural environments. [Method] A dual-branch network integrating Efficient Multi-scale Attention (EMA) and Spatial and Channel Reconstruction Convolution (SCConv) is constructed. Local and global features are extracted through a residual network and an improved dynamic pose-aware Agent Swin Transformer, respectively. SCConv is employed to suppress redundancy, EMA is utilized to enhance multi-scale representation, and head pose priors are incorporated for cross-modal compensation. [Results] On the MPII FaceGaze dataset, the average gaze angular error is reduced to 4.23°; in extreme rotation regions (|Yaw|>15° and |Pitch|>15°), the error is only 5.8°. On the more challenging ETH-XGaze dataset, a promising performance of 4.41° is achieved, demonstrating strong cross-dataset generalization capability. Additionally, the model achieves a single-frame inference latency of only 16.1 ms and an inference frame rate of 62.1 FPS. [Conclusion] This feature fusion method effectively resolves the issues of feature redundancy and scale sensitivity in appearance-based gaze estimation. While maintaining industrial-grade real-time performance, it significantly improves estimation accuracy and robustness under extreme poses and complex scenarios, exhibiting high practical value for engineering applications.

Key words: gaze estimation, feature fusion, EMA module, SCConv module, head pose, robustness