Class-Based Image Comparison for Weather-Aware Abnormality Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for detecting abnormalities in vehicle images fail to accurately account for changes due to weather and time differences when comparing images captured at different times, leading to incorrect detection of changes.
Innovation Solution
An information processing apparatus that analyzes images using class detection, extracts comparison target candidates based on imaging information including date, time, and weather, and selects appropriate images for comparison by considering object types, sizes, and proportions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If images are compared to detect abnormalities, then change detection capability is improved, but false detection increases due to weather and time differences
Solution Approach 1:
The system performs preliminary classification of objects in images before comparison. By detecting and categorizing objects (e.g., vehicles, buildings, vegetation) and their attributes (size, shape, color) in advance, the system can later filter out differences caused by environmental changes and focus only on genuine abnormalities, thereby improving detection accuracy while accounting for weather and time variations.
Solution Approach 2:
The patent applies different processing criteria to different objects based on their local characteristics. Each object class (vehicles, buildings, vegetation) is analyzed with specific features relevant to its nature, allowing the system to distinguish between temporary environmental variations and actual changes in each local context, thus reducing false detections.
2Loss of information
If all past images are stored for comparison, then comprehensive data availability is improved, but storage requirements and processing complexity increase
Solution Approach 1:
The system segments the comparison process by first classifying objects into distinct categories and then performing comparisons only within relevant categories. This segmentation allows the system to manage large datasets more efficiently by processing and storing only the necessary classification information (object type, size, shape, color) rather than treating all image data uniformly, thereby reducing processing complexity while maintaining comprehensive data availability.
Solution Approach 2:
The patent extracts key diagnostic features from images (object classification, size, shape, color) and stores these extracted attributes rather than the complete raw images for all comparisons. This extraction reduces the data volume and processing complexity while preserving the essential information needed for abnormality detection, allowing comprehensive comparison capability with reduced system complexity.
Data Source
AI summary
A class detection means analyzes a captured image captured by using a camera, and classifies objects included in the captured image into a plurality of classes. A candidate extraction means extracts comparison target image candidates from an information accumulation unit based on imaging information of the captured image. An image selection means selects a comparison target image from among the comparison target image candidates based on at least one of types of the classes included in the captured image, sizes of regions of the respective classes, and proportions of the regions of the respective classes with respect to the entire image, and at least one of types of classes included in each of the comparison target image candidates, sizes of regions of the respective classes, and proportions of the regions of the respective classes with respect to the entire image.