基于自校正区域和级联阈值的瞳孔检测方法
A Pupil Detection Method Based on Self-Calibrating Regions and Cascaded Thresholding
摘要
头戴式眼动追踪设备的视线估计精度依赖于瞳孔检测的准确性.针对以往算法在眼妆浓艳、光斑遮挡、瞳孔放大等情况下瞳孔检测准确率及瞳孔定位成功率较低的问题,本文提出了一种基于自校正瞳孔区域与级联阈值分割的瞳孔检测算法.该算法首先通过确定视频首帧中包含瞳孔的区域矩形为后续检测奠定基础,其次结合帧间信息以及小区域间隔采样初步定位暗瞳区域,最后通过自适应光斑补偿、级联阈值分割以及RANSAC 椭圆拟合算法得到最终的瞳孔椭圆轮廓.在基准瞳孔数据集上,与现有最新文献相比,该算法的瞳孔检测准确率提高了 2.83%,瞳孔区域定位成功率提高了 3.24%.实验结果表明:该算法在瞳孔放大、反射光斑遮挡、眼妆浓艳等因素影响下,仍能保持精确的瞳孔轮廓拟合效果.
The accuracy of gaze estimation in head-mounted eye trackers relies on the precision of pupil detection.Conventional methods exhibit lower accuracy in pupil detection and reduced success rates in pupil localization under challenging conditions such as heavy eye makeup,corneal reflection occlusion,and pupil dilation.To address these issues,this paper proposes a pupil detection method combining self-calibrating pupil regions with cascaded threshold segmentation.The approach establishes a baseline by defining the rectangular pupil region from the first frame.Subsequently,it preliminarily locates dark pupil regions through inter-frame information analysis and small-area interval sampling.Finally,adaptive glare compensation,cascaded threshold segmentation,and RANSAC ellipse fitting are employed to determine the optimal pupil contour.Evaluations on benchmark pupil datasets demonstrate that compared with state-of-the-art methods,our approach achieves 2.83%higher pupil detection accuracy and 3.24%improvement in pupil localization success rate.Experimental results confirm that the proposed method maintains precise pupil contour fitting effectiveness despite challenges including pupil dilation,reflective glare occlusion,and heavy eye cosmetics.
年,卷(期):
2026, 65(5)
分类号:
TP391.41
机标分类号:
TP18
TP391.41
TN929.5
在线出版日期:
2026年11月4日