Video Training Model for Shelf Product Action Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing techniques struggle to accurately estimate the object on which a person's action is performed in scenes with many similar objects or complex backgrounds, such as a product shelf in a store.

Innovation Solution

A generation program and method that generates a machine training model by analyzing video data to specify actions of individuals holding products, creating combined image data to train the model to identify the specific object being interacted with, using contrast training to enhance accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a machine training model is trained using conventional methods on image data with many similar objects, then the model can process the data, but the detection accuracy deteriorates

Engineering Contradiction:
Improvedetection accuracyVSAvoidnumber of similar objects
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent segments the training process into two distinct phases: contrast training to differentiate similar objects, and subsequent action detection training. This segmentation allows the model to first learn to distinguish between similar objects before learning action detection, thereby maintaining high accuracy even when many similar objects are present in the scene.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs contrast training as a preliminary action before action detection training. By pre-training the model to distinguish between similar objects using contrastive loss, the model acquires the ability to differentiate objects beforehand, which directly addresses the problem of detection accuracy deterioration in scenes with many similar objects.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If conventional training methods are used without contrast training, then the training process is simpler, but the model cannot accurately distinguish between similar objects

Engineering Contradiction:
Improveobject distinction accuracyVSAvoidtraining process complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent introduces contrast training as a preliminary action that systematically prepares the model for accurate object distinction. Although this adds a training step, it provides a structured and automated approach using contrastive loss functions, which is more efficient than manual feature engineering and yields significant accuracy improvements.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent implements feedback mechanisms through contrastive loss calculation during contrast training. The model receives feedback on its ability to distinguish similar objects, and this feedback is used to iteratively improve the model's feature representations, enabling accurate object distinction even in complex training scenarios.

Inventive Principle:
Principle #23Feedback

3Measurement precision

If action detection is trained without prior contrast training, then the training data requirements are reduced, but the model fails to accurately identify the specific object being interacted with

Engineering Contradiction:
Improveobject identification accuracyVSAvoidtraining time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent performs contrast training as a preliminary action that efficiently prepares the model for action detection. By pre-learning object distinction capabilities through contrastive loss, the model requires less action-specific training data and time, as the foundation for accurate object identification is already established.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent maintains continuity of useful action by seamlessly transitioning from contrast training to action detection training. The model's learned features from contrast training are directly utilized in the subsequent action detection phase, creating a continuous learning process that maximizes training efficiency and minimizes total training time while maintaining high accuracy.

Inventive Principle:
Principle #20Continuity of useful action

Data Source

PatentEP4723023A1Generation program, inference program, generation method, and information processing device
Publication Date: 2026.04.08 FUJITSU LTD
  • EP4723023A1 patent drawingFigure 1
  • EP4723023A1 patent drawingFigure 2
  • EP4723023A1 patent drawingFigure 3

AI summary

A generation program causing a computer to execute a process including acquiring a video obtained by imaging an area including a product shelf on which products are arranged, specifying an action of a person holding the product by analyzing the acquired video, specifying an image frame including a product stored on the product shelf and a product held by the person from a plurality of image frames that form the acquired video based on the specified action of the person holding the product, and generating a machine training model trained to identify a person performing an action of taking out the product from the product shelf using the specified image frame.