Driver Hazard Detection Probability Prediction Using Eye Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems fail to predict the probability of hazard detection by drivers while performing secondary tasks, as they do not account for the duration of forward and non-forward glances and the spillover effect of cognitive load from in-vehicle tasks on primary driving tasks, which increases the risk of accidents.
Innovation Solution
A method and system that track and categorize driver eye movements, using an eye tracker and vehicle data collector to create tuples of forward and non-forward glance durations, and apply transition matrices to estimate the probability of hazard detection, alerting the driver if the probability falls below a threshold.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If drivers perform secondary in-vehicle tasks, then drivers can access infotainment systems and communicate with passengers, but drivers' eyes are taken off the road ahead, impairing hazard detection ability
Solution Approach 1:
The system continuously monitors driver glance behavior and provides real-time feedback by calculating and comparing hazard detection probability against thresholds, alerting drivers when probability falls below safe levels. This closed-loop feedback mechanism dynamically adjusts driver awareness based on actual glance patterns.
Solution Approach 2:
The system performs preliminary analysis of glance sequences and hazard detection probability before actual hazards occur, enabling proactive warnings rather than reactive responses. By predicting future hazard detection capability based on current glance patterns, the system prepares alerts in advance.
2Productivity
If drivers alternate glances between inside of vehicle and forward roadway multiple times during complex tasks, then drivers can complete secondary tasks, but the frequency of glancing increases cognitive load and reduces hazard detection probability
Solution Approach 1:
The system introduces an intermediary computational model that translates complex multi-factor driver behavior data into a single interpretable hazard detection probability metric. This intermediary representation simplifies the relationship between glance alternation frequency and hazard detection capability, enabling more effective monitoring and alerting.
3Measurement precision
If the system monitors and analyzes driver glance sequences in real-time, then the system can predict hazard detection probability, but the computational complexity and data processing requirements increase
Solution Approach 1:
The system segments the continuous stream of driver behavior data into discrete temporal sequences of glances, categorizing them by duration and direction. This segmentation transforms complex continuous data into manageable discrete units that can be processed using sequence modeling techniques, reducing overall computational complexity.
Data Source
AI summary
A system and method for determining a probability that a driver will detect a hazard while driving. An eye tracker tracks the driver's forward and non-forward glance durations. A probability calculator categorizes the glance durations into an observation type and selects a transition matrix specific to that observation type. The probability calculator groups the glance durations and observation type into a tuple and calculates a probability that the driver will detect a hazard in the current time segment based on the tuple, the previous time segment's detection state, and the transition matrix. Based on whether the determined probability is above or below a threshold, the current time segment detection state is stored in memory as a detect state (driver is likely to detect a hazard) or a non-detect state (driver is not likely to detect a hazard). The driver is alerted if determined to be in a non-detect state.