Efficient Pupil Detection Method Based on Multi-stage Image Processing and Hybrid Model Constraints

To enhance the accuracy and robustness of pupil detection, this paper proposes an efficient algorithm that integrates multi-stage image processing and hybrid model constraints. The algorithm employs a binarization method that combines both local and global thresholds, effectively improving the contrast between the pupil and the background while mitigating interference. Additionally, a geometric-symmetric constraint mechanism is introduced to filter out false edge points, thereby improving the precision of pupil edge detection. We also incorporate an ellipse-circle hybrid model for fitting, optimized by a weighted error function and the Huber loss function to enhance robustness against outliers. Experimental results on the Labelled Pupils in the Wild (LPW) dataset demonstrate that, when the detection error is limited to 5 pixels, the detection rate achieves 93%, with an average processing time of 6.0194ms per frame. Practical application tests further validate the algorithm’s robustness and applicability. This algorithm significantly enhances computational efficiency while maintaining high accuracy, offering both substantial theoretical value and promising engineering applications.