LSTM Eye Gaze Prediction Warm-Up Phase
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing eye gaze detection methods face delays and processing limitations, leading to sparse eye gaze position data, which hinders timely determination of user intents and device responses.
Innovation Solution
A Long Short-Term Memory (LSTM) neural network model is used to generate predicted eye gaze positions by processing measured eye gaze positions in a warm-up phase, allowing the model to recalibrate and generate incremental changes, thereby reducing computational effort and improving response time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If eye gaze detection is performed using traditional methods, then measurement precision is achieved, but response time is delayed due to processing limitations
Solution Approach 1:
The system performs a warm-up phase by processing a predetermined number of measured eye gaze positions before generating predictions. This preliminary action prepares the LSTM model's hidden state to capture context-specific gaze patterns, enabling faster and more accurate predictions during the inference phase without sacrificing measurement precision
Solution Approach 2:
The LSTM model generates predicted eye gaze positions that replicate and extrapolate from measured gaze positions. By learning the temporal patterns and incremental changes from historical data, the model creates accurate copies of future gaze positions, reducing the need for continuous real-time processing while maintaining precision
2Loss of time
If continuous eye gaze tracking is performed at high frequency, then response time is improved, but processing resources are overwhelmed
Solution Approach 1:
The system alternates between warm-up phases (processing measured positions) and inference phases (generating predictions). This periodic action allows the model to recalibrate at intervals while providing continuous predictions during inference phases, reducing overall processing load while maintaining responsive performance
Solution Approach 2:
The LSTM model uses its own previous predictions as input for generating subsequent predictions. This self-service mechanism allows the model to maintain temporal consistency and generate accurate predictions without requiring continuous external measured data, reducing processing requirements while improving response time
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
A method includes determining a measured eye gaze position of an eye of a user. The method also includes determining a first incremental change in the measured eye gaze position by processing the measured eye gaze position by a long short-term memory (LSTM) model, and determining a first predicted eye gaze position of the eye at a first future time based on the measured eye gaze position and the first incremental change. The method additionally includes determining a second incremental change in the first predicted eye gaze position by processing the first predicted eye gaze position by the LSTM model, and determining a second predicted eye gaze position of the eye at a second future time subsequent to the first future time based on the first predicted eye gaze position and the second incremental change.