Recognition Dictionary Creation via Quasi-Image Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object recognition technologies in retail checkout systems face challenges in efficiently creating recognition dictionaries for commodities, particularly in handling varying distances and light attenuation, which affects image quality and feature extraction accuracy.
Innovation Solution
A recognition dictionary creation apparatus that includes an image capturing section, measurement module, specifying module, generation module, and registration module, which generates quasi-commodity images at different registration distances and corrects for light attenuation, allowing for efficient registration of feature data across multiple distances.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple reference images are captured at different distances to improve recognition accuracy, then recognition reliability improves, but device complexity and time consumption increase
Solution Approach 1:
The patent creates virtual reference images by copying and transforming a single actual reference image. The image generation unit generates synthetic images at different distances by applying geometric transformations and light attenuation simulations to the captured image, eliminating the need to physically capture multiple images at different distances.
Solution Approach 2:
The patent changes image parameters such as brightness, contrast, and geometric scale to simulate different capture distances. The light attenuation unit adjusts brightness parameters to reflect distance-related lighting changes, while the image generation unit modifies geometric parameters to create the appearance of objects at various distances from the capture device.
2Adaptability or versatility
If reference images are captured at different distances to account for distance variations, then adaptability improves, but measurement precision requirements increase
Solution Approach 1:
Instead of capturing multiple images at precisely controlled distances, the system captures one reference image and generates virtual copies at different distances through image processing. This approach makes the system adaptable to various distances without requiring precise distance measurement and control during image capture.
Solution Approach 2:
The system pre-generates reference images at multiple distances and light conditions before actual recognition occurs. By preparing virtual reference images in advance with varying distance simulations, the system becomes adaptable to different capture distances without requiring real-time distance measurement precision.
3Manufacturing precision
If light attenuation correction is applied to improve feature extraction accuracy, then manufacturing precision of recognition dictionary improves, but processing time increases
Solution Approach 1:
The light attenuation unit applies parameter changes to image brightness and contrast to simulate and correct for light attenuation effects at different distances. By adjusting these visual parameters rather than performing complex physical measurements, the system achieves precise feature extraction while maintaining reasonable processing speeds.
Data Source
AI summary
A recognition dictionary creation apparatus photographs a commodity by an image capturing section to capture the image of the commodity; measures a distance from the image capturing section to the commodity photographed by the image capturing section; generates a quasi-commodity-image in a state in which the commodity is imaginarily moved away from the image capturing section to a given registration distance longer than the measurement distance based on the image of the commodity photographed by the image capturing section; extracts an appearance feature amount of the commodity from the generated quasi-commodity-image; and registers, in a recognition dictionary in association with the registration distance data, the extracted appearance feature amount as the feature amount data of the photographed commodity at the registration distance to create the recognition dictionary.