Conformal Bounding Box Intervals for Object Detector Safety
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detectors in autonomous vehicles lack a reliable method to ensure the safety and accuracy of bounding box coordinate predictions, which is crucial for safe operation.
Innovation Solution
A method and device for determining a prediction interval for bounding box coordinates using conformal quantiles, ensuring the object detector operates safely by providing calibration data, test samples, and conformal label quantiles to establish a reliable safety check.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If object detectors are used in autonomous vehicles, then detection speed and productivity are improved, but reliability and safety of predictions deteriorate due to lack of uncertainty quantification
Solution Approach 1:
The system performs preliminary calibration using calibration data before actual detection operations. Conformal quantiles are pre-computed from calibration data to establish prediction intervals, ensuring reliability is built into the system beforehand rather than added during real-time operation. This allows fast detection while maintaining statistical guarantees.
Solution Approach 2:
The system uses calibration data to compute conformal quantiles that provide feedback on the reliability of predictions. By comparing predicted coordinates against prediction intervals derived from calibration data, the system can determine whether predictions are reliable and adjust operations accordingly, creating a feedback loop that ensures safety.
2Reliability
If prediction intervals are added to object detector output, then reliability is improved, but device complexity increases
Solution Approach 1:
The patent replaces complex mechanical or algorithmic methods for uncertainty estimation with a statistical mathematics-based conformal prediction approach. Instead of using complex neural network ensembles or Bayesian methods, the system uses conformal quantiles computed from calibration data, simplifying the overall system while maintaining reliability.
3Measurement precision
If conformal quantiles are computed for all classes, then measurement precision is improved, but loss of information increases due to selective class filtering
Solution Approach 1:
The system applies different treatments to different classes based on their reliability. High-confidence classes (those in the conformal label set) receive full prediction interval coverage, while low-confidence classes are filtered out. This local differentiation ensures precision is maintained where needed without losing critical class information.
Solution Approach 2:
The system pre-computes conformal label sets that identify which classes have sufficient statistical coverage before making predictions. This preliminary classification allows the system to selectively apply prediction intervals only to reliable classes, maintaining precision while preserving important class information for safe operation.
Data Source
Figure 1~2
Figure 3
AI summary
A device, a data structure, and a method for determining a prediction interval for a coordinate of a bounding box, in particular for checking whether an object detector operates safely or not, preferably for operating an in particular autonomous vehicle, wherein the method comprises providing (202) calibration data and a test sample, wherein the calibration data comprises digital images that are associated with a respective ground truth bounding box coordinate and class label, wherein the test sample comprises a digital image, determining (204), in particular with the object detector, a predicted box coordinate for the box coordinate depending on the digital image of the test sample, determining (206), depending on the calibration data, conformal label quantiles for the respective classes and conformal box coordinate quantiles for the respective box coordinates, selecting (210), depending on the conformal label quantiles, a conformal box coordinate quantile for the box coordinate from the conformal box coordinate quantiles for the box coordinate, determining (212) the conformal box coordinate prediction interval for the box coordinate depending on the conformal box quantile for the box coordinate.