Cluster Analysis for Eye Opening Data Drowsiness Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current drowsiness detection systems for drivers rely on indirect measures and are not robust in accurately distinguishing between normal eye movements and drowsiness or microsleep, leading to potential inaccuracies in warning systems.
Innovation Solution
A method using cluster analysis of eye opening data, including eye opening degree, speed, and acceleration, to classify eye states and determine a robust eye opening level, which improves the detection of drowsiness and microsleep by assigning data sets to specific clusters and averaging weighted measurements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If indirect measures from driving behavior are used to detect drowsiness, then the detection can be implemented without direct eye monitoring, but the measurement precision and reliability of drowsiness detection deteriorates
Solution Approach 1:
The patent replaces indirect mechanical/behavioral inference with direct optical measurement. Video cameras capture actual eye opening states, and image processing algorithms directly measure eyelid positions and eye opening areas, substituting behavioral proxies with direct visual measurement of the physiological parameter itself.
Solution Approach 2:
The patent introduces an intermediary processing system between the eye and the detection result. Image processing algorithms serve as intermediaries that extract precise eye opening measurements from video data, bridging the gap between raw visual input and reliable drowsiness detection metrics.
2Device complexity
If simple eye opening threshold methods are used, then the device complexity is reduced, but the reliability of distinguishing normal eye movements from drowsiness deteriorates
Solution Approach 1:
The patent transforms the detection approach from single-parameter thresholding to multi-parameter analysis. It simultaneously evaluates eye opening area, eyelid position, blink duration, and blink frequency, changing from one-dimensional to multi-dimensional parameter assessment to improve reliability while maintaining computational feasibility.
Solution Approach 2:
The patent adds temporal and dynamic dimensions to the detection system. By analyzing blink duration, frequency, and patterns over time, the system moves beyond static eye opening measurements to incorporate temporal dynamics, enabling reliable distinction between normal blinking and drowsiness-related eye closure.
3Measurement precision
If video camera data processing is used to determine eye opening level, then the measurement precision improves, but the loss of time due to image processing increases
Solution Approach 1:
The patent performs preliminary processing by continuously capturing and pre-processing video frames in real-time. Image processing algorithms continuously analyze eye opening states as video data is captured, preparing measurements in advance so that drowsiness detection can immediately use pre-computed eye opening levels without additional processing delays.
Solution Approach 2:
The patent implements continuous video capture and processing to maintain an ongoing stream of eye opening measurements. The system continuously processes video data to generate real-time eye opening levels, ensuring that detection is based on the most current measurements without time loss from periodic sampling or batch processing.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
The invention relates to a method for classifying eye opening data (208) for at least one eye (202, 242) of an occupant (104) of a vehicle for sensing drowsiness and/or sensing microsleep in the occupant (104). The method comprises a step of generation of a first eye opening data record (220) at a first measurement time in a sliding time window, wherein the first eye opening data record (220) has at least one measurement point (222) that represents a first degree of eye opening and/or a first eyelid speed of movement and/or a first eyelid movement acceleration of the eye (202, 242) of the occupant (104) at the first measurement time, a step of capture of a second eye opening data record (228) at a second measurement time in the sliding time window, wherein the second eye opening data record (228) has at least one capture point (230) that represents a second degree of eye opening and/or a second eyelid speed of movement and/or a second eyelid movement acceleration of the eye (202, 242) of the occupant (104) at the second measurement time, and a step of execution of a cluster analysis using the at least one measurement point (222) and the at least one capture point (230) in order to associate at least the first eye opening data record (220) and/or the second eye opening data record (228) with a first data cluster (236) in order to classify the eye opening data (208), wherein the first data cluster (236) represents an opening state of the eye (202, 242) of the occupant (104).