Depth Training Data Generation Using Synthetic 3D Viewpoints

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional depth sensors for range imaging are expensive and structurally difficult to implement, making the collection of a large set of depth values for different objects in a real environment complex and costly.

Innovation Solution

A system that uses a depth sensor to capture initial images and depth values, combined with a 3D graphics engine to generate synthetic data from different viewpoints, creating a training dataset for a neural network model that includes both real and synthetic data pairs.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If depth sensors are deployed to capture depth values from multiple positions or viewpoints, then the quantity and diversity of depth data increases, but the system complexity and cost increase significantly

Engineering Contradiction:
Improvequantity of depth dataVSAvoidsystem complexity
Core Design Contradiction:
Quantity of substanceVSDevice complexity

Solution Approach 1:

The patent creates synthetic depth images by rendering 3D models from multiple virtual viewpoints, copying the effect of physical depth sensor deployments without the associated complexity. The system generates training data by virtually positioning a camera around a 3D model and synthesizing depth images, thereby obtaining diverse depth data without deploying multiple physical sensors.

Inventive Principle:
Principle #26Copying

2Quantity of substance

If depth sensors are deployed to capture depth values from multiple positions or viewpoints, then the quantity and diversity of depth data increases, but the implementation cost increases

Engineering Contradiction:
Improvequantity of depth dataVSAvoidimplementation cost
Core Design Contradiction:
Quantity of substanceVSEase of manufacture

Solution Approach 1:

The system copies the functional outcome of multiple depth sensor deployments through synthetic data generation. By rendering depth images from virtual viewpoints around a 3D model, the system obtains diverse training data without the cost of purchasing and deploying multiple physical depth sensors.

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The patent replaces expensive, complex depth sensor hardware with computationally-generated synthetic data. The 'cheap' alternative is the software-based synthesis process that generates unlimited training samples without physical hardware costs.

Inventive Principle:
Principle #27Cheap short-living objects (Disposable)

3Measurement precision

If a large set of depth sensors is deployed for collecting depth values, then the quality and coverage of depth measurement improves, but the structural complexity of the system increases

Engineering Contradiction:
Improvedepth measurement qualityVSAvoidsystem structural complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system achieves comprehensive depth measurement coverage by virtually rendering the 3D model from multiple angles and positions. This synthetic approach replicates the measurement quality that would require numerous physical depth sensors, but without the structural complexity of deploying and coordinating multiple sensors.

Inventive Principle:
Principle #26Copying

Data Source

PatentEP4229551B1Training dataset generation for depth measurement
Publication Date: 2026.01.28 SONY GROUP CORP
  • EP4229551B1 patent drawingFigure 1
  • EP4229551B1 patent drawingFigure 2
  • EP4229551B1 patent drawingFigure 3

AI summary

A system for generation of training dataset is provided. The system controls a depth sensor to capture, from a first viewpoint, a first image a first depth value associated with the first object. The system receives tracking information from a handheld device associated with the depth sensor, based on a movement of the handheld device and the depth sensor in a 3D space. The system generates graphic information corresponding to the first object based on the received tracking information. The graphic information includes the first object from a second viewpoint. The system calculates a second depth value associated with the first object, based on the graphic information. The system generates, for a neural network model, a training dataset which includes a first combination of the first image and the first depth value, and a second combination of second images corresponding to the graphic information and the second depth value.