Autonomous Vehicle Error Detection for Dynamic Object Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous vehicle systems lack effective methods to automatically detect and store data associated with dynamic object recognition errors, particularly in imbalanced data sets, leading to potential detection failures.
Innovation Solution
An apparatus and method for autonomously detecting dynamic object recognition errors by parsing sensor data, analyzing it with a dynamic object recognition model, and storing relevant data in a non-volatile memory when specific error conditions are met, such as a high ratio of certain objects or low confidence scores, using a processor-controlled system with sensors, a parser, and error detection logic.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If a dynamic object recognition model learns from imbalanced data sets where specific objects are underrepresented, then the model can be trained with available data, but detection performance on underrepresented objects deteriorates with high detection errors
Solution Approach 1:
The system performs preliminary detection of detection errors during the testing phase before training. By identifying underrepresented objects with high error rates in advance, the system prepares targeted training data augmentation strategies, allowing the model to learn from corrected and augmented data in subsequent training iterations, thereby improving detection performance on previously problematic objects
Solution Approach 2:
The system implements a feedback loop where detection results from the dynamic object recognition model are continuously evaluated against ground truth data. Errors are automatically detected and fed back into the training process through data augmentation, creating an iterative improvement cycle that progressively enhances detection performance for underrepresented objects
2Measurement precision
If manual detection and storage of error-prone data is performed, then detection performance can be improved through selective learning, but time consumption and labor costs increase significantly
Solution Approach 1:
The system implements automated self-detection of recognition errors by comparing model outputs with ground truth data. The error detection module automatically identifies underrepresented objects with high error rates and triggers data augmentation processes without human intervention, eliminating manual labor and significantly reducing time consumption while maintaining the ability to improve detection performance through selective learning
Solution Approach 2:
The system replaces manual mechanical processes of error detection and data selection with automated computational algorithms. The error detection module uses computational comparisons between predicted and actual data to automatically identify problematic cases, substituting human analysts with algorithmic processes that are faster, more consistent, and scalable
3Quantity of substance
If all sensor data is stored for analysis, then comprehensive training data is available, but storage requirements and processing complexity increase significantly
Solution Approach 1:
The system extracts only the specific subset of data that is most valuable for improving model performance - namely, data involving underrepresented objects with high detection error rates. By selectively extracting and storing only these error-prone cases rather than all sensor data, the system reduces storage requirements and processing complexity while still providing sufficient training material to improve detection performance
Data Source
AI summary
A method of automatically detecting a dynamic object recognition error in an autonomous vehicle is provided. The method includes parsing sensor data obtained by frame units from a sensor device equipped in an autonomous vehicle to generate raw data by using a parser, analyzing the raw data to output a dynamic object detection result by using a dynamic object recognition model, determining that detection of a dynamic object recognition error succeeds by using an error detector when the dynamic object detection result satisfies an error detection condition, and storing the raw data and the dynamic object detection result by using a non-volatile memory when the detection of the dynamic object recognition error succeeds.