Inference Program for Object Detection Using Difference Masks

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing self-checkout systems struggle to detect objects that do not exist in the background image but are present in the target image, particularly when dealing with unknown objects or multiple objects close to each other.

Innovation Solution

An inference program that uses a combination of feature extraction models, generation models, and estimation models to generate mask images and specify the regions of objects that do not exist in the background image but are present in the target image.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a simple object detection method is used, then the device complexity is low, but the detection precision deteriorates when dealing with unknown objects or multiple objects close to each other

Engineering Contradiction:

Improveobject detection precisionVSAvoidmodel structure complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the object detection task into three distinct stages: (1) background image processing to extract background features, (2) target image processing to extract target features and generate difference masks, and (3) combined feature processing to detect objects. This segmentation allows each stage to specialize in specific aspects, improving overall detection precision while managing complexity through modular architecture.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent introduces an intermediate difference mask as a mediator between the target image and the final detection result. This mask highlights regions where objects appear in the target image but not in the background, serving as an intermediary feature that guides the detection process and improves precision for unknown or closely positioned objects.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Measurement precision

If multiple processing stages are used to improve detection accuracy, then the detection precision improves, but the processing time increases

Engineering Contradiction:

Improveobject detection precisionVSAvoidprocessing time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent performs preliminary processing by extracting and storing background image features before the actual detection occurs. This preliminary action prepares the system in advance, allowing the detection stage to focus only on identifying differences between the target and background, thereby reducing the computational burden during real-time detection while maintaining high precision.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent extracts only the essential difference regions using a difference mask, separating the relevant object information from the entire image. This extraction approach reduces the amount of data that needs to be processed in subsequent stages, decreasing processing time while preserving detection precision by focusing computational resources on relevant regions.

Inventive Principle:

Principle #2Taking out (Extraction)

3Measurement precision

If feature extraction and mask generation are performed, then the object detection accuracy improves, but the computational load increases

Engineering Contradiction:

Improveobject detection accuracyVSAvoidcomputational energy consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent applies local quality by generating difference masks that highlight only the specific regions where objects appear, rather than processing the entire image uniformly. This localized approach concentrates computational energy on relevant regions, improving detection accuracy while reducing overall computational load by ignoring background areas that do not contain objects.

Inventive Principle:

Principle #3Local quality

Data Source

PatentUS12236674B2Computer-readable recording medium storing inference program, computer-readable recording medium storing learning program, inference method, and learning method

Publication Date: 2025.02.25 FUJITSU LTD

AI summary

A non-transitory computer-readable recording medium stores an inference program for causing a computer to execute processing including: acquiring a background image in which a target area in which an object is arranged is captured, and a target image in which the object and the area are captured; generating an intermediate feature by inputting the background image and the target image to a feature extraction model; generating a mask image that indicates a region of an object that does not exist in the background image but exists in the target image by inputting the intermediate feature to a generation model; and specifying the object that does not exist in the background image but exists in the target image by inputting the generated mask image and intermediate feature to an estimation model.