Eye center annotation is vital for ophthalmic diagnostics and surgery. However, existing algorithms often require specialized equipment and face challenges in real-time performance, particularly under varying lighting. This study evaluates four widely used facial landmarking algorithms—Mediapipe, Dlib, Haar Cascade, and RetinaFace—in the task of eye iris center annotation. The optimal algorithm is employed to validate the effectiveness in optokinetic nystagmus (OKN) detection and eyesight assessment. The results demonstrate that Mediapipe outperforms the other algorithms, offering superior real-time performance, high accuracy, and robust adaptability to different lighting conditions. Additionally, this study validates its potential in eyesight detection.
References
[1]
Zhang, Y. and Li, X. (2020) Face Detection Using Deep Learning: A Survey. Computer Vision and Image Understanding, 191, Article 102871.
[2]
Falkenstein, I.A., Cochran, D.E., Azen, S.P., Dustin, L., Tammewar, A.M., Kozak, I., et al. (2008) Comparison of Visual Acuity in Macular Degeneration Patients Measured with Snellen and Early Treatment Diabetic Retinopathy Study Charts. Ophthalmology, 115, 319-323. https://doi.org/10.1016/j.ophtha.2007.05.028
[3]
Suh, D.W. and Shahraki, K. (2023) Vision Screening Claims for Young Children in the United States. Pediatrics, 152, e2023062804. https://doi.org/10.1542/peds.2023-062804
[4]
Ambrosino, C., Dai, X., Antonio Aguirre, B. and Collins, M.E. (2023) Pediatric and School-Age Vision Screening in the United States: Rationale, Components, and Future Directions. Children, 10, Article 490. https://doi.org/10.3390/children10030490
[5]
Bailey, I.L. and Lovie-Kitchin, J.E. (2013) Visual Acuity Testing. from the Laboratory to the Clinic. VisionResearch, 90, 2-9. https://doi.org/10.1016/j.visres.2013.05.004
[6]
US Preventive Services Task Force (2017) Vision Screening in Children Aged 6 Months to 5 Years: US Preventive Services Task Force Recommendation Statement. Journal of the American Medical Association, 318, 836-844.
[7]
Garcia, F. and Soto, R. (2021) Enhancements of Mediapipe for Real-Time Eye Tracking and Gaze Estimation. Journal of Computer Vision, 59, 129-142.
[8]
Liao, M. and Wang, H. (2019) Efficient Real-Time Eye Tracking Using Haar Cascades and Deep Learning. Vision Technology, 52, 1124-1135.
[9]
Gupta, S. and Roy, D. (2020) Real-Time Multi-Face and Eye Detection with Dlib and Open CV. In: Proceedings of the International Conference on Computer Vision, Springer, 45-50.
[10]
Wu, P. and Zhang, H. (2021) Retina Face: A Practical Single-Stage Dense Face Localization in the Wild.
[11]
Aigbe, S. and Zhang, Z. (2022) Improving Eye Center Annotation Accuracy in Real-Time Systems Using Mediapipe. Journal of Machine Learning Research, 23, 111-123.
[12]
King, D.E. (2009) Dlib-ML: A Machine Learning Toolkit. Journal of Artificial Intelligence Research, 2, 1-6.
[13]
Sahoo, B. and Li, L. (2020) Challenges and Improvements in Facial Landmark Detection for Robust Eye Center Annotation. IEEE Transactions on Image Processing, 29, 7845-7857.
