Road Feature Extraction via Global-Local Model Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for extracting enhanced quality road features from data, such as satellite and aerial images, face challenges due to complex ground features, occlusions, shadows, and inconsistencies in orthorectification processes, leading to fragmented and incomplete road boundary detection.
Innovation Solution
A system that uses a combination of a global feature segmentation model and a local feature detection model to determine break points and generate cropped feature maps, allowing for accurate and complete extraction of road features by preserving topology and improving connectivity and completeness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional deep learning based semantic segmentation models are used to detect road features, then high recall and precision rates are achieved, but topology correctness deteriorates especially at road intersections or merges/splits
Solution Approach 1:
The patent segments the road feature detection process into two distinct stages: (1) a global feature segmentation model that performs semantic segmentation to identify road features with high precision, and (2) a local feature detection model that specifically targets topological features like intersections, merges, and splits. This segmentation allows each model to specialize in its strength while the results are integrated to achieve both detection precision and topology correctness.
Solution Approach 2:
The patent merges the outputs of the global feature segmentation model and the local feature detection model to produce the final road feature detection results. The global model provides accurate road boundary detection while the local model corrects and enhances topological features, combining their strengths to resolve the contradiction between detection precision and topology correctness.
2Manufacturing precision
If hand-crafted feature extraction and curve-fitting processes are used to detect road features, then limited circumstances are handled well, but productivity deteriorates due to the large amount of working time required
Solution Approach 1:
The patent replaces the mechanical hand-crafted feature extraction and curve-fitting processes with deep learning-based automated models. The global feature segmentation model and local feature detection model automatically learn and extract road features from satellite and aerial images, eliminating the need for manual feature engineering and significantly improving processing speed while maintaining or enhancing accuracy.
Solution Approach 2:
The patent transforms the feature extraction process from manual parameter-based methods to data-driven learning-based methods. By training the deep learning models on large datasets, the system automatically adapts to various road scenarios and environmental conditions, achieving both high accuracy and improved productivity compared to traditional hand-crafted approaches.
3Quantity of substance
If conventional methods are used to extract road features from satellite and aerial images, then some regional road boundary completion is achieved, but measurement precision deteriorates due to occlusion from vehicles, vegetation, constructions and shadow cast from trees and buildings
Solution Approach 1:
The patent applies preliminary action by first using the global feature segmentation model to detect road features across the entire image, then using the local feature detection model to specifically address and correct features in occluded or shadowed regions. This two-stage approach ensures comprehensive road boundary completion while maintaining precision even in challenging environmental conditions.
Solution Approach 2:
The local feature detection model acts as an intermediary that refines and corrects the output of the global feature segmentation model, particularly in regions affected by occlusion and shadow. It mediates between the initial detection results and the final accurate road feature extraction, improving measurement precision in difficult-to-detect areas.
Data Source
AI summary
An approach is provided for generating feature data. The approach, for example, involves determining a set of break points associated with at least one road feature based on mask image data. The mask image data is associated with overhead image data comprising the at least one road feature. The approach further involves generating a set of cropped feature maps based on processing of a global feature map obtained from a global feature segmentation model and the set of break points. The global feature map is associated with the overhead image data. The approach further involves generating the feature data associated with the at least one road feature based on application of a local feature detection model on the generated set of cropped feature maps and the set of break points. The approach further involves storing the generated feature data in a geographic database.