Eye Movement Data Encoding for Automated Fixation and Saccade Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing eye tracking technologies struggle to accurately and automatically classify eye movement data into fixations, saccades, and smooth pursuits, especially in dynamic scenes, requiring user expertise and parameterization, and fail to provide intuitive and robust analysis for end-users.
Innovation Solution
A method that encodes eye movement and eye tracking data using time and space parameters, applying anisotropic filtering and threshold-based filtering to reduce data volume, enabling a ternary classification of fixations, saccades, and smooth pursuits without user parameterization, and using machine learning for data interpretation and automation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional algorithms with user-defined parameters are used for eye movement classification, then the analysis can be adapted to specific applications, but the method becomes too complex and requires user expertise that is rarely available
Solution Approach 1:
The system performs self-parameterization by automatically determining optimal parameters for fixation detection, saccade detection, and smooth pursuit detection based on the input data characteristics. The algorithm adapts to different applications and subjects without requiring external user input, making the system both versatile and easy to use.
Solution Approach 2:
The patent implements dynamic parameter adjustment where the algorithm automatically modifies detection parameters based on data characteristics. Instead of fixed user-defined thresholds, the system adapts parameters such as velocity thresholds, acceleration thresholds, and spatial dispersion criteria to match the specific experiment conditions and subject characteristics.
2Measurement precision
If high sampling rate eye tracking data is recorded to improve measurement precision, then the time-resolved position data becomes more accurate, but the data volume increases significantly
Solution Approach 1:
The patent extracts only the essential information from high-volume raw eye tracking data by identifying and isolating key events (fixations, saccades, smooth pursuits). Instead of processing all raw data points, the algorithm extracts meaningful patterns and represents them in a compact format, significantly reducing data volume while preserving measurement precision.
Solution Approach 2:
The continuous stream of high-frequency eye tracking data is segmented into discrete eye movement events. By dividing the data into meaningful units (individual fixations, saccades, and smooth pursuit segments), the system reduces the overall data volume while maintaining the precision of each detected event through the ternary classification approach.
3Adaptability or versatility
If conventional algorithms are used to separate smooth pursuit movements, then the analysis can handle dynamic scenes, but the separation is currently regarded as difficult and has not been satisfactorily solved
Solution Approach 1:
The patent implements a dynamic detection approach that adapts to different types of eye movements based on their characteristic patterns. The algorithm dynamically adjusts detection criteria for smooth pursuit by analyzing velocity, acceleration, and spatial continuity patterns, enabling reliable separation of smooth pursuit from fixations and saccades in dynamic scenes.
Solution Approach 2:
The system uses feedback from the detected eye movement patterns to continuously refine the classification. By analyzing the characteristics of detected movements and adjusting detection parameters accordingly, the algorithm improves the reliability of smooth pursuit detection, particularly in dynamic scenes where the viewer follows moving objects.
Data Source
AI summary
An apparatus and method of encoding eye movements and eye tracking data (DAT), represented as time and space parameters (t, x, y) obtained by an eye tracking device and assigned each to a viewpoint (A, B, C, D, E . . . ). Pairs of numbers (Z0,Z1; Z0,Z2; Z0,Z3 . . . ) are taken from the groups of numbers and are combined with each other to obtain for each combination a value (W) that indicates at least a spatial distance (S) between two viewpoints (E, C), wherein the obtained values (W) represent an encoding of the eye movement and eye tracking data (DAT). Preferably the values (W) are determined and stored in form of a first matrix (M) or array. The matrix is subjected to one or more operations (smooth filtering, anisotropic filtering, threshold filtering, anisotropic diffusion) such that the resulting matrix represents an encoding of fixations, saccades and smooth pursuit of a raw data scanpath.