Cross-View Image Matching Using Class Distribution Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for ground-to-aerial cross-view matching, such as those using CNNs, do not consider extracting features from images other than RGB images or their orientation maps, limiting their ability to accurately match ground-view and aerial-view images.
Innovation Solution
The proposed solution involves acquiring ground-view and aerial-view images, along with class information indicating the distribution of classes of objects, and extracting features from these images and class information to compute ground, aerial, and class features. These features are then used to determine whether the ground-view and aerial-view images match each other.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If only RGB images and orientation maps are used for feature extraction, then the system complexity is kept simple, but the matching accuracy between ground-view and aerial-view images is limited
Solution Approach 1:
The patent segments the feature extraction process into multiple independent components: RGB image feature extraction, orientation map feature extraction, and class information feature extraction. Each component processes a specific type of data independently, allowing the system to incorporate multiple feature sources without creating complex interdependencies, thus improving matching accuracy while managing system complexity
Solution Approach 2:
The patent creates a universal feature extraction framework that can process multiple types of input data (RGB images, orientation maps, and class information) through a common neural network architecture. This multi-functional approach allows the system to leverage diverse data sources for matching while maintaining a unified and manageable system structure
2Measurement precision
If class information is extracted and processed, then the matching accuracy improves through additional semantic features, but the computational time and processing complexity increase
Solution Approach 1:
The patent performs preliminary extraction of class information from images before the main matching process. By pre-processing and extracting semantic class features in advance, the system prepares additional matching criteria without adding computational burden during the actual matching operation, thus improving accuracy while minimizing time loss
Solution Approach 2:
The patent uses class information as an additional feature representation that parallels the visual features. Instead of complex processing of raw image data, the system creates a simplified class-level abstraction that captures semantic information efficiently, reducing computational time while maintaining high matching accuracy
Data Source
AI summary
An image matching apparatus acquires a ground-view image, an aerial-view image, and class information. The class information indicates a distribution of classes of objects on the ground-view image, the aerial-view image, or both. The image matching apparatus-extracts features from the ground-view image to compute a ground image feature, extracts features from the aerial-view image to compute an aerial image feature, and extracts features from the class information to compute a class feature. The image matching apparatus determines whether or not the ground-view image and the aerial-view image match each other based on the ground image feature, the aerial image feature, and the class feature.