Synthetic 3D Image Generation for Object Recognition Training

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional image generation techniques for object recognition systems are limited in providing sufficient image properties to identify objects across a wide range of situations, particularly when objects are outside the training set or in different environmental conditions.

Innovation Solution

The development of an image generating apparatus and method that creates a three-dimensional representation of objects with associated properties, using data from computer-aided drawing models, sensor information, and material composition, to produce training data for object recognition algorithms, enabling identification from various distances, angles, and using multiple sensors.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If conventional image generation techniques using public photographs are used, then the training data can be obtained easily, but the object recognition system cannot identify objects outside the training set or in different environmental conditions

Engineering Contradiction:

Improveobject recognition capability across different conditionsVSAvoidrecognition accuracy for objects outside training set

Core Design Contradiction:

Adaptability or versatilityVSReliability

Solution Approach 1:

The patent creates synthetic copies of objects through 3D modeling and rendering instead of using only photographic copies. This allows generating unlimited training variations of objects under different conditions (lighting, angles, environments) without needing actual physical objects or photographs, thereby improving adaptability while maintaining recognition reliability

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The system varies multiple parameters including lighting conditions, camera angles, object positions, environmental backgrounds, and sensor types when generating synthetic training images. This parameter variation enables the object recognition system to learn robust features that generalize across different real-world conditions, resolving the contradiction between adaptability and reliability

Inventive Principle:

Principle #35Parameter changes

2Reliability

If three-dimensional representations with multiple properties are generated synthetically, then object recognition accuracy improves across diverse conditions, but the data processing and generation complexity increases

Engineering Contradiction:

Improverecognition accuracyVSAvoidprocessing system complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The synthetic data generation system serves multiple functions: it creates training data for various object types, simulates different sensor modalities (visible light, infrared, radar), generates multiple viewing angles and lighting conditions, and produces data for diverse environmental scenarios. This multi-functionality improves recognition reliability across conditions while avoiding the need for separate systems for each function, thereby managing complexity

Inventive Principle:

Principle #6Universality (Multi-functionality)

3Adaptability or versatility

If synthetic training data with comprehensive properties is used, then object identification from various distances and angles improves, but the time and computational resources required for generation increase

Engineering Contradiction:

Improveidentification capability from various distances and anglesVSAvoiddata generation time

Core Design Contradiction:

Adaptability or versatilityVSLoss of time

Solution Approach 1:

The system pre-generates comprehensive synthetic training data covering multiple distances, angles, lighting conditions, and environmental scenarios before the object recognition system is deployed. This preliminary action creates a robust training dataset that enables the system to quickly identify objects in various real-world conditions without requiring time-consuming data collection or processing during actual operation

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The synthetic data generation process can continuously produce training samples across different parameters (angles, distances, lighting) without interruption or repetition. This continuous generation capability ensures comprehensive coverage of identification scenarios while optimizing computational resource utilization, balancing adaptability improvement with time efficiency

Inventive Principle:

Principle #20Continuity of useful action

Data Source

PatentUS10931931B2Apparatus, system, and method for generating an image

Publication Date: 2021.02.23 THE BOEING CO

AI summary

Described herein is a method of generating an image that includes receiving a set of data corresponding to an object. The method also includes generating a three-dimensional representation of the object using the set of data. The method includes generating properties for the object using the set of data. The method also includes associating the properties with the three-dimensional representation of the object, wherein the three-dimensional representation of the object and the properties for the object are used to produce training data for an object recognition algorithm.