| 39 | 0 | 239 |
| 下载次数 | 被引频次 | 阅读次数 |
为解决瞳孔中心定位算法在光照变化、遮挡及个体差异下鲁棒性不足、耗时较长的问题，提出联合ROI(region of interest)检测与Daugman算法优化的轻量化瞳孔定位算法。构建以GE＿ShV2为骨干网络的YOLOv8n模型，通过结构优化降低46.91%计算量，模型体积和参数分别减少43.55%与46.18%，同时保持99%的检测精度；在颈部网络引入C2f＿ESCA模块实现通道稀疏注意力加权，在DySample基础上新增候选点重要性加权抽样，根据噪声可靠度与边缘梯度动态分配采样概率，有效增强边缘感知，减少漏采。对Daugman算法引入灰度差平方和与积分图加速算法，将复杂度从O(N)降至O(1)，有效提升定位效率与抗干扰性，缩短定位时间。实验结果表明，改进后的算法在处理边缘细节和光照变化时表现更加稳健。
Abstract：To address the issues of insufficient robustness and long time consumption of the pupil center localization algorithm under varying illumination, occlusion, and individual differences, a lightweight pupil localisation algorithm combining ROI(region of interest) detection and Daugman optimisation was proposed. The YOLOv8n model with GE＿ShV2 as the backbone network was constructed to reduce 46. 91% of the computation volume by structural optimization, the model volume and parameters were reduced by 43. 55% and 46. 18% respectively, while 99% of the detection accuracy was maintained. The C2f＿ESCA module was introduced into the neck network to achieve the sparse attention weighting of the channels, and the new candidate points were added on the basis of DySample. Importance-weighted sampling was added on the basis of DySample, and the sampling probability was dynamically allocated according to the noise reliability and edge gradient, which effectively enhances edge perception and reduces leakage. The grey level difference sum of squares calculation method and integral graph acceleration algorithm were introduced into the Daugman algorithm, leading to the lowered complexity from O(N) to O(1), improved positioning efficiency and anti-interference, and shortened positioning time. The experimental results show that the improved algorithm performs more robustly when dealing with edge details and light changes.
[1]苟超，卓莹，王康，等.眼动跟踪研究进展与展望[J].自动化学报，2022, 48(5):1173-1192.GOU C, ZHUO Y, WANG K, et al. Research advances and prospects of eye tracking[J]. Acta Automatica Sinica, 2022, 48(5):1173-1192.(in Chinese)
[2] BONTEANU P, BOZOMITU R G, CRACAN A,et al. A new pupil detection algorithm based on multiple angular integral projection functions[C]∥Proceedings of the 2021 International Conference on eHealth and Bioengineering. Piscataway:IEEE Press,2021:1-4.
[3] FRASCA M, LA TORRE D. An extended Daugman’s algorithm for iris with eye pathology recognition[J].Expert Systems with Applications, 2024, 252:124160.
[4]马启玮，韩慧妍，韩燮，等.基于星射线法的瞳孔中心定位[J].计算机工程与设计，2021, 42(5):1409-1417.MA Q W, HAN H Y, HAN Y, et al. Pupil center location based on star ray method[J]. Computer Engineering and Design, 2021, 42(5):1409-1417.(in Chinese)
[5]于哲舟，刘岩，刘元宁.基于YOLOV3改进的虹膜定位算法[J].东北大学学报（自然科学版），2022,43(4):496-500, 508.YU Z Z, LIU Y, LIU Y N. Improved iris locating algorithm based on YOLOV3[J]. Journal of Northeastern University(Natural Science), 2022, 43(4):496-500, 508.(in Chinese)
[6]杨亚男，朱晓冬，刘元宁，等.基于改进YoloV4网络的虹膜定位算法[J].吉林大学学报（理学版），2022,60(2):369-380.YANG Y N, ZHU X D, LIU Y N, et al. Iris location algorithm based on improved YoloV4 network[J].Journal of Jilin University(Science Edition), 2022,60(2):369-380.(in Chinese)
[7]于青禾，张田，刘怀钦.一种改进YOLOv7的轻量化二次虹膜定位算法[J].微电子学与计算机，2024,41(8):40-52.YU Q H, ZHANG T, LIU H Q. A lightweight secondary iris localization algorithm with improved YOLOv7[J]. Microelectronics&Computer,2024,41(8):40-52.(in Chinese)
[8] XUE K J, WANG J S, WANG H. Research on pupil center localization detection algorithm with improved YOLOv8[J]. Applied Sciences, 2024, 14(15):6661.
[9]李洪波，何启学，鲍胜利.基于改进YOLOv8n的轻量级入侵物种检测算法[J/OL].计算机应用，1-13,[2025-05-27]. https：∥kns. cnki. net/kcms/detail/51.1307. tp. 20250401. 1550. 016. html.LI H B, HE Q X, BAO S L. Lightweight invasive species detection algorithm based on improved YOLOv8n[J/OL]. Journal of Computer Applications, 1-13,[2025-05-27]. https：∥kns. cnki. net/kcms/detail/51.1307. tp. 20250401. 1550. 016. html.(in Chinese)
[10]孙语，刘文龙，蒋茂松.基于注意力机制和空洞卷积的瞳孔定位算法[J].电子测量技术，2023, 46(15):126-132.SUN Y, LIU W L, JIANG M S. Pupil location algorithm based on Attention Gate and dilated convolution[J]. Electronic Measurement Technology, 2023,46(15):126-132.(in Chinese)
[11] WANG Q L, WU B G, ZHU P F,et al. ECA-Net:efficient channel attention for deep convolutional neural networks[C]∥2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR).[S.l.]:IEEE, 2020:11531-11539.
[12]刚帅，刘培胜，郭希旺.改进基于YOLOv8n的轻量化钢材表面缺陷检测算法[J].电子测量技术，2025,48(3):74-82.GANG S, LIU P S, GUO X W. Lightweight improved YOLOv8n model for steel defect detection features[J]. Electronic Measurement Techonology,2025, 48(3):74-82.(in Chinese)
[13] TONG Z J, CHEN Y H, XU Z W, et al. Wise-IoU:bounding box regression loss with dynamic focusing mechanism[EB/OL].[2024-11-04]. https：//arxiv.org/pdf/2301. 10051.
[14] LIU W Z, LU H, FU H T, et al. Learning to upsample by learning to sample[C]∥Proceedings of the 2023 IEEE/CVF International Conference on Computer Vision. Piscataway:IEEE Press, 2023:6004-6014.
[15] LI H L, LI J, WEI H B. et al. Slim-neck by GSConv:a lightweight-design for real-time detector architectures[J]. Journal of Real-Time Image Processing, 2024, 21. DOI:org/10. 1007/s11554-024-01436-6.
[16] SWATHI A, AARTI&KUMAR S. A smart application to detect pupil for small dataset with low illumination[J]. Innovations in Systems and Software Engineering, 2021, 17:29-43.
[17] CHENG G, CHAO P Z, YANG J, et al. SGSTYOLOv8:an improved lightweight YOLOv8 for realtime target detection for campus surveillance[J].Applied Sciences, 2024, 14(12):5341.
[18]张奇，陈梦蝶，赵杰.融合特征金字塔和通道注意力的轻量车辆检测算法[J].国外电子测量技术，2023,42(1):41-48.ZHANG Q, CHEN M D, ZHAO J. Lightweight vehicle detection network fusing feature pyramid and channel attention[J]. Foreign Electronic Measurement Technology, 2023, 42(1):41-48.(in Chinese)
[19] TAN M X, LE Q V. EfficientNetV2:smaller models and faster training[EB/OL].[2024-11-04], https：∥arxiv. org/abs/2104. 00298.
基本信息:
中图分类号:TP391.41;TP18
引用信息:
[1]武丽,丁琴.ROI引导的Daugman瞳孔定位算法[J].中国科技论文,2025,20(10):872-884.