ROI引导的Daugman瞳孔定位算法
ROI-guided Daugman pupil localisation method
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。
计算技术、计算机技术
瞳孔检测YOLOv8轻量化Daugman瞳孔定位
丁琴,武丽.ROI引导的Daugman瞳孔定位算法[EB/OL].(2025-10-31)[2025-11-02].http://www.paper.edu.cn/releasepaper/content/202510-38.点此复制
n order to solve the defects of the pupil centre localisation algorithm, which is susceptible to poor robustness and long time-consumption under light, occlusion and individual differences, a lightweight pupil localisation algorithm combining ROI detection and Daugman optimisation is proposed. The YOLOv8n model with GE_ShV2 as the backbone network is constructed to reduce 46.91% of the computation volume by structural optimization, and the model volume and parameters are reduced by 43.55% and 46.18% respectively, while maintaining 99% of the detection accuracy; the C2f_ESCA module is introduced into the neck network to achieve the sparse attention weighting of the channels, and the new candidate points are added on the basis of DySample. Importance-weighted sampling is added on the basis of DySample, and the sampling probability is dynamically allocated according to the noise reliability and edge gradient, which effectively enhances edge perception and reduces leakage. The Daugman algorithm introduces the grey level difference sum of squares calculation method and integral graph acceleration algorithm, which reduces the complexity from O(N) to O(1), effectively improves the positioning efficiency and anti-interference, and shortens the positioning time. The experimental results show that the improved algorithm performs more robustly when dealing with edge details and light changes.
pupil detectionYOLOv8lightweightingdaugmanpupil localisation
展开英文信息
评论