Pupil accurate detection method based on rapid human eye semantic segmentation network

Abstract

The invention discloses a pupil accurate measurement method based on a rapid human eye semantic segmentation network. The method comprises the steps: (a) carrying out the accurate segmentation of a pupil based on the rapid human eye semantic segmentation network, so as to obtain a pupil region; (b) in the fast human eye semantic segmentation network, expanding convolution and common convolution are in parallel, and the local refinement capability is improved while the receptive field is expanded; and (c) introducing a plurality of attention generation modules, fully extracting semantic features, and obtaining a refined semantic result. Features of different levels are extracted through down-sampling, and reasoning is carried out by combining up-sampling to the size of an original image. According to the method, refined reconstruction of pupils is completed on the basis of pupil semantic feature extraction, so that automatic semantic segmentation is carried out. Some blind searches existing in previous algorithms are avoided.

Description

technical field [0001] The invention relates to the field of detection, in particular to an accurate pupil detection method based on a fast human eye semantic segmentation network. Background technique [0002] In the field of computer vision research, it has always been one of the main research directions in this field to carry out human eye-related technology research by extracting human eye features. As an important intermediate link in human eye detection, pupil detection has important application prospects in blink detection, fatigue detection, human-computer interaction and other fields. [0003] Someone proposed a pupil detection algorithm (Hough-Contour) which combines Hough circle transformation and contour matching. For each frame of image, first perform grayscale and filter denoising; then extract the edge and use the modified Hough gradient method to detect the initial circle as the pupil parameter; finally use the position and radius near the pupil on the filte...

AI Technical Summary

Problems solved by technology

However, since the distance between the pupil and the upper and lower eyelids is very similar, simply deepening the U-Net network may lead to unclear boundary segmentation, so this method is not suitable for semantic segmentation of the pupil

[0006] To sum up, the current research focus is gradually shifting from traditional image processing algorithms to deep learning, but there is still no end-to-end learning pupil semantic segmentation algorithm, although the existing traditional image processing pupil detection methods have achieved fruitful research results , but the demand for high precision, high robustness and high generalization ability of pupil detection is still an urgent problem to be solved

Images