Learning Image Generation Apparatus for Natural Object Motion

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing technologies fail to effectively generate learning images configured with continuous images that reduce unnatural object motion, which is essential for recognizing behaviors in videos.

Innovation Solution

A learning image generation apparatus and method that acquires background and object continuous images, determines synthesis positions based on camera posture information, and synthesizes objects with the background images to generate continuous images with natural object motion.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If object images are synthesized onto background images to generate learning continuous images, then the quantity of learning images increases, but unnatural object motion occurs in the generated continuous images

Engineering Contradiction:

Improvenumber of learning imagesVSAvoidnaturalness of object motion

Core Design Contradiction:

Quantity of substanceVSManufacturing precision

Solution Approach 1:

The patent introduces camera posture information as an intermediary element to mediate between the object continuous image and the background image. By using posture information (position, angle, zoom) as a mediator, the system can accurately transform object images to match the background camera's viewpoint, thereby reducing unnatural motion while generating learning continuous images.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent applies parameter changes by transforming object images based on camera posture parameters (position coordinates, angle, zoom ratio). By dynamically adjusting these parameters according to the background camera's posture, the system generates synthesized images that maintain natural object motion characteristics throughout the continuous image sequence.

Inventive Principle:

Principle #35Parameter changes

2Manufacturing precision

If camera posture information is used to determine synthesis positions, then object motion naturalness improves, but the device complexity increases

Engineering Contradiction:

Improvenaturalness of object motionVSAvoidcomplexity of image generation system

Core Design Contradiction:

Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent applies preliminary action by acquiring and storing camera posture information in advance before generating the learning continuous images. By pre-obtaining the background camera's posture data (position, angle, zoom), the system can efficiently determine synthesis positions without complex real-time calculations, thereby reducing overall system complexity while maintaining motion naturalness.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS11429814B2Learning image generation apparatus, learning image generation method, and non-transitory storage medium

Publication Date: 2022.08.30 NEC CORP

AI summary

The present invention provides a learning image generation apparatus (100) including: a background image acquisition unit (111) that acquires a background image; a background camera posture information acquisition unit (112) that acquires posture information of a background camera which generates a background image; an object continuous image acquisition unit (113) that acquires an object continuous image including an object; a synthesis position determination unit (121) that determines a synthesis position on the background image of the object included in each of a plurality of object still images included in the object continuous image based on the posture information of the background camera; and an image synthesis unit (131) that synthesizes the objects included in each of the plurality of object still images with the background image to generate a synthesis continuous image based on the background image, the object continuous image, the synthesis position.