In today’s digital world, most real-world applications like computational photography and augmented reality effects depend on estimating the eye position by tracking the iris. The accurate iris tracking can then be used to determine the metric distance from the camera to the user. This is done even without using a dedicated depth sensor by the use of machine learning.
Google has announced the release of MediaPipe Iris for accurate iris estimation. This machine learning model is based on MediaPipe Face Mesh. MediaPipe Iris (GitHub) can track the iris, pupil, and the eye contours without the need for any specialized hardware. A single RGB camera present in the device does the thing in real-time. Then this model uses these Iris Landmarks to determine the metric distance between the subject and the camera pretty accurately.
Is MediaPipe Iris Accurate & Safe
As per the Google Blog, MediaPipe Iris measures the metric distance with a relative error of less than 10% without the use of a depth sensor.
Google has also clarified that the iris tracking does not provide any form of identity recognition. The system uses MediaPipe, an open-source framework developed by Google for researchers and developers to build world-class ML applications. It runs on most modern mobile phones, laptops, desktops, and even on the web.
Depth Estimation from a Single Image
Google boasts that its iris-detection model determines the distance of the camera to a subject with less than 10% error. The fact that the horizontal iris diameter remains roughly constant at 11.7±0.5 mm across a wide range of population, the model exploits this value along with some other geometric arguments. Using the EXIF data of the image, the ML model uses different parameters to determine the distance of the person from the camera.
MediaPipe Iris Release
Google is releasing the iris and depth estimation models as a cross-platform MediaPipe pipeline. This means that it can run on all devices — desktop, mobile, and the web. The Iris ML pipeline will run locally in the browser, without sending any data to the cloud. This guarantees the privacy of the users and makes sure the user has full control over his data.