Background Image Trapezoid Correction for Learning Data Generation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The challenge is to generate learning data for an identification model by superimposing a detection target image on a background image, where the background image is captured at an angle that does not face the detection target's visible surface, making direct superimposition impossible.

Innovation Solution

The solution involves performing projective transformation and trapezoid correction on the background image to align its projection method with that of the detection target image, followed by superimposing the detection target image on the corrected background image to generate composite learning data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Area of stationary object

If the background image is captured at an angle that does not face the detection target's visible surface, then the imaging device can capture the fixture and vicinity, but the detection target image cannot be simply superimposed on the background image

Engineering Contradiction:

Improvecoverage area of background imageVSAvoidalignment precision between images

Core Design Contradiction:

Area of stationary objectVSManufacturing precision

Solution Approach 1:

The patent applies projective transformation to change the geometric parameters of the background image, transforming it from the original capture angle to match the detection target image's perspective. This parameter transformation enables accurate superimposition while maintaining the utility of angle-varied background images for diverse learning scenarios

Inventive Principle:

Principle #35Parameter changes

2Measurement precision

If manual annotation is performed to generate learning data, then accurate label information can be obtained, but the generation process becomes troublesome and time-consuming

Engineering Contradiction:

Improveaccuracy of label informationVSAvoidtime for learning data generation

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent uses automated image processing (projective transformation and superimposition) to generate learning data from existing images, replacing manual annotation processes. This copying approach creates accurate learning data by transforming and combining existing images rather than manually creating each annotation from scratch

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent performs projective transformation and image superimposition in advance to pre-generate learning data. This preliminary action creates ready-to-use learning datasets that can be directly applied for training identification models, eliminating the need for time-consuming manual annotation during deployment

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentEP3742384B1Information processing apparatus, article identification apparatus, and article identification system

Publication Date: 2025.04.16 TOSHIBA TEC KK

Figure 1
Figure 2
Figure 3

AI summary

According to one embodiment, an information processing apparatus includes a memory and a processor. The memory stores a background image that does not include a detection target, a detection target image that is an image of the detection target, and label information that indicates the detection target of the detection target image. The processor performs trapezoid correction on the background image, generates a composite image by superimposing the detection target image on the background image subjected to the trapezoid correction, and generates learning data based on the composite image and the label information.