(Subscribe to our Today's Cache newsletter for a quick snapshot of top 5 tech stories. Click here to subscribe for free.)
Google has released a new solution for real-time iris tracking and depth estimation on mobile devices using machine learning.
The machine learning model – MediaPipe Iris, is designed for accurate iris estimation. It can track landmarks involving the iris, pupil and eye contours using a single RGB camera, in real-time, without the need for specialised hardware, Google said in a blog post.
The company further explains that due to the limited computing capacity on mobile devices and other factors, including variable light conditions and people squinting, iris tracking can be challenging.
Mobile devices use hardware like dedicated depth sensor to assess distance between camera and user for real-world applications of iris tracking, including computational photography like portrait mode on a smartphone camera; and augmented reality effects in apps and games. These applications rely on estimating eye position by tracking the iris.
The process involves using an earlier MediaPipe Face Mesh model to generate a mesh of the approximate face geometry. The eye region is then isolated for use in the MediaPipe Iris model.
About 50 thousand images, representing a variety of illumination conditions and head poses from geographically diverse regions were used to train the model, Google said.
Left: MediaPipe Iris predicting metric distance in cm, from iris tracking alone, without the use of a depth sensor. Right: Ground-truth depth.
The search giant claims the MediaPipe Iris model is able to assess the distance between the user and the camera with relative error of less than 10% from facial landmarks by using the focal length of the camera.
The model also relies on the fact that the horizontal iris diameter of the human eye remains roughly constant at 11.7mm (±0.5) across a wide population, it added.
MediaPipe Iris can run on most modern mobile phones, desktops, laptops and even on the web, because of the open source cross-platform framework, and can be accessed by researchers and developers.
