Vehicle Camera Recognition Feedback for Vulnerability-Based Retraining
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for training image recognition models in autonomous vehicles lack consideration for the performance of the recognition network, leading to inadequate selection and cultivation of training data, resulting in suboptimal object recognition accuracy.
Innovation Solution
A vehicle system that includes a sensor part with multiple cameras and a controller capable of training an image recognition model and a vulnerability assessment model using convolutional neural networks, which assesses the vulnerability of the image recognition model by determining a vulnerability score based on intersection over union and object recognition results, and adjusts the model accordingly to improve object recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual scene code classification is used for training data selection, then operation simplicity is maintained, but object recognition accuracy deteriorates due to inadequate training data cultivation
Solution Approach 1:
The system enables automatic training data selection and model retraining without manual intervention. The vulnerability assessment model automatically identifies vulnerable training data, and the image recognition model is retrained using this selected data, making the system self-improving and eliminating manual operation requirements while enhancing object recognition accuracy
Solution Approach 2:
The system implements a feedback mechanism where the vulnerability assessment model evaluates the image recognition model's performance, identifies vulnerable training data based on vulnerability scores, and feeds this information back for model retraining. This closed-loop feedback continuously improves object recognition accuracy by addressing weaknesses automatically
2Measurement precision
If vulnerability assessment model is trained to automatically assess image recognition model performance, then object recognition accuracy is improved through continuous remediation, but device complexity increases
Solution Approach 1:
The system segments the overall model assessment and improvement process into distinct functional modules: an image recognition model for object detection, a vulnerability assessment model for evaluating performance and identifying weak points, and a training data selection mechanism. This segmentation allows each component to specialize and work together to improve accuracy while managing complexity through modular design
Solution Approach 2:
The vulnerability assessment model is pre-trained to assess the image recognition model's performance before actual deployment. By performing preliminary assessment and identifying vulnerable training data in advance, the system can proactively retrain the image recognition model to prevent accuracy degradation, thereby improving object recognition performance while organizing complexity through structured preparation
3Productivity
If real-time training data selection is implemented without performance consideration, then productivity is improved through automated selection, but manufacturing precision deteriorates due to inadequate training data quality
Solution Approach 1:
The system replaces manual mechanical selection of training data with an automated vulnerability assessment mechanism. The vulnerability assessment model automatically evaluates training data quality by analyzing vulnerability scores, identifying which data samples are most beneficial for model improvement. This substitution maintains high productivity through automation while ensuring training data quality by using objective vulnerability-based criteria rather than arbitrary selection
Data Source
AI summary
A vehicle may include: a sensor part including a plurality of cameras having fields of view different from each other; and a controller configured to process image data obtained by the sensor part, wherein the controller is configured to: train an image recognition model outputting a feature map by inputting training data stored in a learning database to the image recognition model, train a vulnerability assessment model outputting a vulnerability score by inputting a feature map output from the trained image recognition model to the vulnerability assessment model, extract a feature map by inputting the image data obtained by the sensor part to the trained image recognition model, determine a vulnerability score by inputting the extracted feature map to the trained vulnerability assessment model, determine whether logging is required based on the vulnerability score, and based on a determination that logging is required, store logging data in the learning database.