Vision System Classification Stability via Confidence Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vision systems for occupant restraint systems face challenges in accurately classifying objects in vehicle interiors due to temporary changes in image data, leading to potential improper classification switching and inappropriate airbag deployment.
Innovation Solution
A method that adjusts classification confidence levels and time periods based on vehicle operating conditions, using a vision system with a classifier that determines image classes and filters classifications based on confidence levels and time periods, ensuring robust and accurate classification by locking onto determined classes until new conditions meet predetermined confidence and time thresholds.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the vision system uses a statistical classifier to determine image classes, then the classification capability is improved, but the system becomes susceptible to temporary changes in image data causing improper classification switching
Solution Approach 1:
The system performs preliminary classification to determine an initial image class, then requires sustained confidence over a time period before allowing class switching. This preliminary action filters out temporary fluctuations in image data that would otherwise cause improper classification changes.
Solution Approach 2:
The system dynamically adjusts the confidence level threshold and time period requirements based on vehicle operating conditions. When conditions suggest potential temporary changes (e.g., door opening, ignition changes), the system increases the confidence and time requirements, making classification switching more difficult and thus more stable.
2Speed
If the system allows frequent class switching in response to image data changes, then the responsiveness to real changes is improved, but false class switching increases leading to inappropriate airbag deployment
Solution Approach 1:
Before allowing class switching, the system performs preliminary verification by checking if the new classification maintains confidence above a threshold for a specified time period. This preliminary check ensures that only sustained, real changes trigger class switching, filtering out false positives while maintaining response to genuine events.
Solution Approach 2:
The system continuously monitors classification confidence levels and compares them against thresholds over time periods. This feedback mechanism allows the system to distinguish between temporary fluctuations and sustained changes, enabling rapid response to real changes while preventing false switching.
3Device complexity
If the system uses fixed confidence level thresholds for class switching, then the control logic is simple, but it cannot adapt to different vehicle operating conditions affecting classification reliability
Solution Approach 1:
The system makes the confidence level threshold and time period parameters dynamic rather than fixed. These parameters are adjusted based on vehicle operating conditions such as door opening status, ignition state, and vehicle speed. This dynamic adaptation allows the system to maintain appropriate classification stability across different operating scenarios without requiring completely different control logic for each condition.
Solution Approach 2:
The system changes the parameters (confidence threshold and time period) based on vehicle operating conditions. When conditions indicate potential temporary changes in the vehicle environment, the system increases these parameters to require higher confidence and longer sustained periods before allowing class switching, thereby adapting to different operational contexts.
Data Source
AI summary
A system for classifying image data includes a vision system (58) for continuously imaging an area (42) of interest and providing an image signals. A classifier (84) determines an image class corresponding to each image signal, determines a confidence level for each determined image class, determines a time period for the confidence level, and establishes an image classification in response thereto. Switching between image classifications is in response to the confidence level and time period of the confidence level. When used in a vehicle restraining system (20), other monitor vehicle parameters (54) can be used to adjust the confidence levels and time periods needed to switch between image classifications.