ADAS Neural Network Retraining for Corner Case Labeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Advanced driver assistance systems (ADAS) face challenges in accurately detecting corner cases due to insufficient training data, leading to low performance in rare events and scenarios, which are labor-intensive and costly to label.
Innovation Solution
A method involving filtering sensor data, such as camera images and lidar point clouds, to identify and label corner cases, allowing for retraining of neural network models using diverse and high-quality datasets, potentially eliminating the need for simulation data and reducing the cost of data collection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If neural network models are trained using comprehensive sensor data including all images, then the model coverage increases, but the labeling cost and time increase significantly
Solution Approach 1:
The patent extracts and isolates only the corner case images from the comprehensive sensor data using automated detection algorithms. Instead of labeling all images, the system identifies and extracts specifically those images depicting corner cases (rare events, edge scenarios), thereby reducing labeling time while maintaining model coverage for critical scenarios.
Solution Approach 2:
The system performs self-service by automatically detecting and filtering corner case images without requiring manual review of all images. The automated detection mechanism enables the system to identify corner cases independently, reducing the need for human annotators to examine every image while still capturing all relevant corner cases for training.
2Measurement precision
If neural network models are retrained using diverse corner case data, then the detection accuracy for rare events improves, but the data collection and processing complexity increases
Solution Approach 1:
The patent segments the training data into two distinct categories: common cases and corner cases. By dividing the comprehensive sensor data into these segments and applying different processing approaches (automated filtering for corner cases, selective labeling), the system improves detection accuracy for rare events while managing processing complexity through structured data organization and specialized handling procedures.
3Manufacturing precision
If all sensor data is labeled for retraining, then the training data quality improves, but the cost and resource consumption increase
Solution Approach 1:
The patent applies local quality by providing high-quality labeled data specifically for corner case images while using automated detection for common cases. Instead of uniformly labeling all images with the same level of effort, the system concentrates labeling resources on corner cases where human judgment is most valuable, thereby improving training data quality for critical scenarios while reducing overall resource consumption.
Data Source
AI summary
A computer-implemented method comprises: receiving sensor data collected during operation of a vehicle, the sensor data including images from at least one camera on the vehicle and other sensor data from at least one other sensor on the vehicle; identifying, based on the sensor data, a corner case for a first neural network model of an advanced driver assistance system (ADAS) of the vehicle; filtering the images to identify a subset of the images depicting the corner case, the filtering performed using a second neural network model; labeling the subset of the images with regard to the corner case, wherein a remainder of the images are not labeled; and retraining the first neural network model using the labeled subset of the images and a subset of the other sensor data corresponding to the subset of the images.