Probabilistic Graphical Model for Gaze Tracking Error Reduction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current gaze tracking systems suffer from inaccuracies due to noise in eye-tracker data, head movements, and other environmental factors, making it difficult to reliably use gaze information for object tracking and interaction.

Innovation Solution

A probabilistic-based process is employed to predict user fixation patterns in displayed scenes by combining eye-tracker data with prior knowledge about the scene, using a probabilistic graphical model to improve the accuracy and robustness of gaze-contingent interaction.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Speed

If gaze tracking data is used directly for object tracking, then the system responds quickly to user input, but the tracking accuracy deteriorates due to noise in eye-tracker data

Engineering Contradiction:
Improveresponse speedVSAvoidtracking accuracy
Core Design Contradiction:
SpeedVSMeasurement precision

Solution Approach 1:

The system performs preliminary actions by predicting future gaze positions based on historical gaze data and scene information before the actual gaze occurs. This allows the system to compensate for noise in real-time gaze data while maintaining quick response, as the prediction is prepared in advance using probabilistic models that incorporate scene context and object movement patterns.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent introduces an intermediary probabilistic prediction system between the raw gaze data and the object tracking output. This intermediary layer filters noise by combining multiple sources of information (historical gaze data, scene description, object positions) to generate a more accurate estimated gaze position, thereby improving tracking accuracy without sacrificing response speed.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Measurement precision

If traditional filtering methods are applied to reduce noise in gaze data, then measurement precision improves, but the system complexity increases

Engineering Contradiction:
Improvegaze data accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The probabilistic prediction system serves multiple functions simultaneously: it filters noise from gaze data, predicts future gaze positions, incorporates scene context, and tracks moving objects. This multi-functionality allows the system to improve measurement precision without proportionally increasing complexity, as a single integrated model performs what would otherwise require multiple separate processing stages.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The system changes the parameters of the prediction model dynamically based on scene characteristics and object properties. By adjusting model parameters such as prediction horizon, noise variance, and scene context weightings, the system adapts to different conditions without requiring fundamentally different processing architectures, thereby managing complexity while maintaining precision.

Inventive Principle:
Principle #35Parameter changes

3Reliability

If the system tracks only large objects, then tracking reliability is maintained, but the applicability to small objects deteriorates

Engineering Contradiction:
Improvetracking reliabilityVSAvoidobject size range
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The system performs preliminary prediction of gaze positions and object interactions before actual tracking occurs. This advance prediction allows the system to maintain reliability for small objects by anticipating their positions based on scene context and movement patterns, rather than relying solely on real-time gaze data which is too noisy for small target detection.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The probabilistic prediction system acts as an intermediary that bridges the gap between noisy gaze data and small object tracking. By generating predicted gaze positions that incorporate scene context and object priors, this intermediary layer enables reliable tracking of small objects that would otherwise be indistinguishable from noise in direct gaze tracking.

Inventive Principle:
Principle #24Intermediary (Mediator)

4Measurement precision

If probabilistic prediction is used to improve tracking accuracy, then measurement precision improves, but the computational requirements increase

Engineering Contradiction:
Improvetracking accuracyVSAvoidcomputational energy
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The system applies partial probabilistic prediction by selectively modeling only the most relevant aspects of gaze behavior and scene context. Rather than performing exhaustive probabilistic calculations for all possible factors, the system focuses on key predictors such as recent gaze history and salient scene elements, achieving improved accuracy with reduced computational energy consumption.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP2992405B1System and method for probabilistic object tracking over time
Publication Date: 2025.02.12 MIRAMETRIX INC
  • EP2992405B1 patent drawingFigure 1a
  • EP2992405B1 patent drawingFigure 1b
  • EP2992405B1 patent drawingFigure 1c

AI summary

A system and method are provided for object tracking in a scene over time. The method comprises obtaining tracking data from a tracking device, the tracking data comprising information associated with at least one point of interest being tracked; obtaining position data from a scene information provider, the scene being associated with a plurality of targets, the position data corresponding to targets in the scene; applying a probabilistic graphical model to the tracking data and the target data to predict a target of interest associated with an entity being tracked; and performing at least one of: using the target of interest to determine a refined point of interest; and outputting at least one of the refined point of interest and the target of interest.