Eye Tracking Calibration Using Polygon Grid Interpolation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional eye tracking systems for Near-Eye-Display (NED) devices are computationally intensive during calibration, consuming significant resources and impacting performance on compact systems, and struggle to accurately track user gaze in real-time, leading to potential motion sickness and suboptimal user experience.

Innovation Solution

A user-specific calibration method is implemented where a NED device presents virtual stimuli randomly or pseudo-randomly to capture eye tracking data, forming a polygon grid from calibration ellipse centers to interpolate real-time gaze direction, using proportionality factors based on distances within the grid to determine the user's optical axis.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional eye tracking calibration is performed using refined anatomical models, then measurement precision of user gaze is improved, but use of energy and computational resources increases significantly

Engineering Contradiction:
Improvegaze tracking accuracyVSAvoidcomputational resource consumption
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The calibration process is segmented into two distinct phases: an initial comprehensive calibration phase that establishes a lookup table, and a real-time operation phase that uses pre-computed data. This segmentation allows the computationally intensive model refinement to occur only once during calibration, while real-time gaze tracking uses lightweight interpolation queries on the pre-built lookup table, dramatically reducing ongoing computational resource consumption.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system performs preliminary calibration actions before real-time operation by pre-computing the anatomical model and storing calibration data in a lookup table. This preliminary action captures all the computationally intensive processing needed for model refinement, so that during actual use, the system only needs to perform simple data retrieval and interpolation operations, minimizing real-time energy consumption.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If real-time gaze tracking is performed using refined anatomical models, then measurement precision is improved, but device complexity increases

Engineering Contradiction:
Improvegaze tracking accuracyVSAvoidcomputational system complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system segments computational tasks between calibration time and runtime. During calibration, complex anatomical modeling is performed offline to create a lookup table. During runtime, only simple interpolation operations are needed, reducing real-time computational complexity while preserving measurement precision through the use of pre-computed anatomical data.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Instead of performing complex anatomical model calculations in real-time, the system creates a simplified copy or representation of the calibrated data in the form of a lookup table. This copy contains all necessary calibration information in a format that enables fast, simple interpolation operations during real-time gaze tracking, reducing device complexity while maintaining accuracy.

Inventive Principle:
Principle #26Copying

3Measurement precision

If calibration data is collected using multiple virtual stimuli, then measurement precision of ellipse centers is improved, but loss of time during calibration increases

Engineering Contradiction:
Improveellipse center accuracyVSAvoidcalibration duration
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system merges multiple calibration data points collected from different virtual stimuli into a single comprehensive lookup table. By combining information from multiple stimuli presentations into one integrated calibration structure, the system achieves high measurement precision through aggregated data while enabling efficient real-time operation through unified data access, effectively managing the trade-off between calibration thoroughness and operational efficiency.

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentEP3853693B1User-specific eye tracking calibration for near-eye-display (NED) devices
Publication Date: 2024.08.14 MICROSOFT TECHNOLOGY LICENSING LLC
  • EP3853693B1 patent drawingFigure 1~2
  • EP3853693B1 patent drawingFigure 3~4
  • EP3853693B1 patent drawingFigure 5A

AI summary

Technologies for performing user-specific calibration of eye tracking systems for Near-Eye- Display (NED) devices. The NED device may sequentially present different virtual stimuli to a user while concurrently capturing instances of eye tracking data. The eye tracking data reveals calibration ellipse centers that uniquely correspond to individual virtual stimuli. The calibration ellipse centers may be used define a polygon grid in association with a sensor plane. The resulting polygon grid is used during operation to interpolate the real-time gaze direction of the user. For example, a real-time instance of eye tracking data may be analyzed to determine which particular polygon of the polygon grid a real-time ellipse center falls within. Then, distances between the real-time ellipse center and the vertices of the particular polygon may be determined. A proportionality factor is then determined based on these distances and is used to interpolate the real-time eye gaze of the user.