Data-Based Model Validation via Reference Classification Comparison
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Data-based models used in safety-critical applications like driver assistance systems require validation to ensure accurate object classification, but existing methods lack a reliable and efficient way to validate these models across various distances and scenarios.
Innovation Solution
A computer-implemented method and device that validate data-based models by determining a measure of confidence based on digital signals from sensors like radar or LIDAR, comparing classifications with a reference model, and storing relevant data pairs to create a well-suited training dataset, allowing for reliable validation and potential retraining when necessary.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If data-based models are used for object classification in driver assistance systems, then classification accuracy can be improved, but validation reliability becomes insufficient without proper validation methods
Solution Approach 1:
The patent applies preliminary action by creating a validation method that performs classification tests on training data before deploying the data-based model in safety-critical applications. The system预先 validates the model by comparing its classifications against reference classifications from established models, ensuring reliability before actual use in driver assistance systems
Solution Approach 2:
The patent uses an intermediary approach by introducing a reference model as a mediator between the data-based model and the validation process. The reference model provides reference classifications that serve as an intermediate standard for evaluating whether the data-based model meets required reliability thresholds
2Reliability
If validation is performed across various distances and scenarios, then validation thoroughness is improved, but computational complexity and time increase
Solution Approach 1:
The patent applies partial action by performing validation on a representative subset of training data covering key distance ranges and scenarios rather than exhaustively testing every possible condition. This allows thorough validation of critical performance aspects while limiting time consumption to acceptable levels
Solution Approach 2:
The patent segments the validation process by dividing it into distance-based categories (e.g., short distance, medium distance, long distance) and scenario types. This segmentation allows systematic validation across various conditions while organizing the computational workload into manageable segments that can be processed efficiently
3Measurement precision
If a reference model is used for comparison, then classification correctness can be verified, but device complexity increases
Solution Approach 1:
The patent uses copying by creating a reference model that replicates or mirrors the structure and functionality of the data-based model being validated. This reference model serves as a simplified copy that provides reference classifications for comparison, enabling correctness verification without requiring overly complex validation infrastructure
Data Source
AI summary
A device, a memory medium, a computer program, and a computer-implemented method for validating a data-based model for classifying an object into a class for an object type or a function type for a driver assistance system of a vehicle. The classification is determined as a function of a digital signal using the data-based model. A reference classification for the object is determined as a function of the digital signal, using a reference model. It is checked, as a function of the classification and the reference classification, whether or not the classification of the data-based model for the object is correct, and the data-based model is validated or not validated, depending on whether or not the classification is correct. The classification and the reference classification are determined for a set of digital signals that are associated with different distances between the object and a reference point.