Video eye pupil positioning and tracking method and device

Abstract

The invention provides a video eye pupil positioning and tracking method and device. The method comprises the following steps: acquiring a current frame of eye image Hi, and inputting previous frames of eye images Hi-1 and Hi into an adversarial network to obtain a predicted next frame of eye image Hi + 1; inputting the Hi + 1 into the first segmentation network to obtain a predicted segmentation mask Yi + 1 of the eye pupil of the next frame; inputting Hi and Yi into a second segmentation network to obtain an eye pupil segmentation mask of the current frame; and repeating the steps until all frames of eye images are processed. According to the method, the adversarial network is arranged, and the relevance of the two adjacent frames of eye images is utilized to predict the next frame of eye image, so that the predicted position of the eye pupil segmentation mask of the next frame is obtained, and the noise influence can be effectively removed. According to the method, the automatic tracking of the eye pupil can be realized only by labeling the two frames of images at the beginning by utilizing the correlation of the two adjacent frames of images, so that the labeling quantity is greatly reduced.

Description

technical field [0001] The invention belongs to the technical field of video tracking, and in particular relates to a video pupil positioning tracking method and device. Background technique [0002] Eye pupil positioning not only plays an important role in safe driving detection, but also plays a vital role in VR (Virtual Reality, virtual reality), AR (Augmented Reality, augmented reality) and other fields. Eye pupil positioning generally includes the following three steps: face detection, human eye region positioning, and human eye pupil positioning. [0003] The existing pupil positioning methods generally use traditional image processing methods, such as hough transform method, ellipse fitting method and gradient vector method. Although these traditional image processing methods have certain advantages in processing speed, their accuracy is not satisfactory; especially when the human eye area is affected by illumination changes and occlusions, it is difficult for tradit...

AI Technical Summary

Problems solved by technology

Since this method simply uses the feature learning ability of the deep neural network, the eye pupil positioning model is established based on the training of a single picture, and the correlation information between the upper and lower frames of the video is not considered. Therefore, although the accuracy of the algorithm has been improved, the real-time performance, resistance to object occlusion, etc. need to be improved, and a large number of image annotations are required

Images