Driver Concentration Estimation Using Autoencoded Gaze Signals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in accurately estimating a driver's concentration level during driving due to difficulties in modeling the causal relation between biological information and internal states, and varying driving conditions, leading to inconsistent attention observation.
Innovation Solution
An electronic device employing an autoencoder architecture with an encoder and decoder to estimate a driver's concentration level by analyzing line-of-sight data, incorporating subject attributes and internal state information, and adjusting parameters based on reproducibility and distribution deviation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional methods calculate cumulative visibility by comparing line of sight with nearby objects, then driver attention can be monitored, but the measurement precision is insufficient due to inability to accurately model causal relations between biological information and internal states
Solution Approach 1:
The patent introduces an autoencoder model as an intermediary computational system that processes multiple biological information indicators (line of sight, pupil diameter, blinking frequency) along with driving conditions and vehicle information. This intermediary model learns complex nonlinear relationships and causal connections between these diverse inputs and the driver's concentration level, achieving high measurement precision without requiring explicit complex modeling by the user
Solution Approach 2:
The system dynamically adjusts the weightings and parameters of different biological information indicators based on driving conditions and vehicle states. The autoencoder learns optimal parameter configurations for different scenarios, allowing the measurement precision to be maintained across varying driving conditions while adapting the model parameters automatically
2Adaptability or versatility
If the system monitors driver attention using biological information, then driver safety can be improved, but the adaptability to diverse driving conditions is reduced due to inconsistent attention observation
Solution Approach 1:
The autoencoder model is designed with multi-functionality to handle diverse inputs including various biological information indicators, driving conditions, and vehicle information. This universal model structure allows it to adapt to different driving scenarios while maintaining consistent and reliable attention observation across all conditions
Solution Approach 2:
The system dynamically adapts its processing based on real-time driving conditions and vehicle states. The autoencoder learns to adjust its internal representations and output interpretations according to the current context, enabling reliable attention measurement that is adaptive to varying driving environments
3Measurement precision
If multiple biological information indicators are collected to improve estimation accuracy, then measurement precision increases, but the device complexity increases due to need for processing multiple data streams
Solution Approach 1:
The patent merges multiple biological information indicators (line of sight, pupil diameter, blinking frequency) along with driving conditions and vehicle information into a unified autoencoder model. This consolidation processes all data streams simultaneously through learned feature interactions, achieving high measurement precision while managing complexity through integrated processing rather than separate analysis of each indicator
Data Source
AI summary
An electronic device includes an encoder and a decoder. The encoder is configured to estimate an unknown value on the basis of first biological information including a line of sight of a subject extracted from an image of the subject, subject's attribute information representing an attribute of the subject, and subject's internal state information representing an internal state of the subject. The decoder is configured to estimate second biological information including the line of sight of the subject on the basis of the unknown value, the subject's attribute information, and the subject's internal state information. The electronic device adjusts parameters of the encoder and the decoder on the basis of reproducibility of the first biological information from the second biological information.