Driver Conduct Inference Using Feature Point Transit Regions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional conduct inference apparatuses in vehicles can only detect specific conducts after they have been performed, preventing early warning systems from alerting drivers to potentially unsafe actions.
Innovation Solution
A conduct inference apparatus that acquires images of drivers, detects feature points, and uses pre-defined conduct inference models to anticipate specific conducts by identifying return points, transit regions, and conduct directions, allowing for the inference of intended actions before they are completed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the conventional conduct inference apparatus detects specific conducts only after they have been performed, then the detection method is simple, but the ability to provide early warning is lost
Solution Approach 1:
The system performs preliminary detection by identifying preparatory movements (hand moving toward face, mouth, or ear) before the specific conduct is completed. The conduct inference model includes a transit region that captures these preparatory phases, allowing the system to warn drivers before they actually perform dangerous actions like using a cellular phone or smoking.
Solution Approach 2:
The detection space is segmented into multiple regions: a return point (where the feature point is located when performing the specific conduct), a transit region (which the feature point passes through when intending to perform the conduct), and other regions. This segmentation allows the system to distinguish between preparatory movements and actual conduct performance.
2Measurement precision
If the system uses a simple hand position detection method, then the device complexity is low, but false alarms increase and detection accuracy decreases
Solution Approach 1:
The system requires detection of the feature point in the transit region for a predetermined time period before triggering an alarm. This preliminary detection phase filters out transient or accidental movements, reducing false alarms while maintaining simple detection logic.
Solution Approach 2:
The system dynamically tracks the movement of the feature point over time, detecting whether it passes through the transit region toward the return point. This temporal dimension adds precision without requiring complex static analysis, as the system evaluates the trajectory and duration of movements.
3Loss of time
If the system detects conduct only after completion, then the response time is short, but the safety warning time is insufficient
Solution Approach 1:
The system detects preparatory movements in the transit region before the driver completes the specific conduct. By identifying hand movements toward the face, mouth, or ear during the transit phase, the system provides advance warning while maintaining rapid response, as the detection threshold is triggered during the preparatory phase rather than waiting for completion.
Data Source
AI summary
In a conduct inference process, feature points are extracted from a capture image. The extracted feature points are collated with conduct inference models to select conduct inference models in each of which an accordance ratio between a target vector and a movement vector is within a tolerance. Among the selected conduct inference models, one conduct inference model in which a distance from a relative feature point to a return point is shortest is selected. Then, a specific conduct designated in the selected conduct inference model is tentatively determined as a specific conduct the driver intends to perform. Furthermore, based on the tentatively determined specific conduct, it is determined whether the specific conduct is probable. When it is determined that the specific conduct is probable, an alarm process is executed to output an alarm to the driver.