Object Detection Augmentation for Autonomous Vehicle Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection systems in autonomous vehicles face challenges in accurately identifying and classifying objects, especially under unusual environmental conditions or when objects are partially occluded, due to the rarity of such scenarios in training data.
Innovation Solution
The system augments training data by flipping or modifying frames and objects to create more diverse and representative scenarios, including rare objects and occluded conditions, thereby enhancing the model's ability to detect and classify objects accurately.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If training data is collected from real-world environments, then the model learns from actual operational conditions, but rare scenarios such as unusual environmental conditions or partially occluded objects are insufficiently represented
Solution Approach 1:
The patent creates synthetic copies of existing training data by applying geometric transformations (flipping, rotating, cropping) and visual modifications (adding occlusions, changing lighting conditions) to generate augmented training samples. This copying approach allows the model to learn from diverse scenarios without requiring additional real-world data collection, directly addressing the scarcity of rare scenarios in real-world training datasets
Solution Approach 2:
The patent systematically varies parameters of training images including orientation (rotation angles), position (translation), scale (cropping ratios), and visual properties (occlusion levels, lighting conditions). By changing these parameters to generate augmented samples, the model exposure to a broader range of scenarios is increased while maintaining the underlying semantic content, thereby improving detection accuracy for rare conditions
2Productivity
If the model is trained on limited diverse scenarios, then training data collection is efficient, but the model fails to accurately detect objects under unusual environmental conditions or partial occlusions
Solution Approach 1:
The patent performs data augmentation during the training data preparation phase rather than during inference. By pre-generating augmented training samples with various occlusions, rotations, and environmental conditions before model training begins, the system efficiently prepares comprehensive training data that covers rare scenarios, enabling the model to learn robust object detection without sacrificing training efficiency
Solution Approach 2:
The patent generates multiple synthetic copies of each training image with different transformations and occlusion patterns. This creates a expanded training dataset that maintains the efficiency of using existing data while significantly increasing the diversity of scenarios the model encounters during training, thereby improving classification accuracy for edge cases
Data Source
AI summary
Techniques for improving training data used to train a model to detect objects in an environment are disclosed. Portions of a dataset that are associated with significant operational conditions may be associated with indications of such conditions. The properties of these portions may be used to identify other portions of the dataset or other datasets for augmentation. Data in these portions of a dataset representing detections in an environment may be augmented generally, for example, by flipping geometric values and modifying associated data accordingly. Specific objects represented in such portions may also be augmented by determining particular objects that are represented in selected frames and channels and modifying the pixel values for pixels associated with such objects. The pixel values may be modified or zeroed based on various criteria. The resulting augmented data may be used to train a machine-learned model to more accurately detect objects in an environment.