Vehicle Neural Network Synthetic Data Augmentation

Solution Overview

Problem

Training deep neural networks for autonomous vehicle systems requires large, time-consuming, and expensive datasets to cover all possible trailer appearances and environmental conditions, making it costly and inefficient to determine trailer angles accurately.

Innovation Solution

The method involves generating synthetic images using photorealistic rendering software and generative adversarial networks to augment training datasets, allowing the network to learn from a smaller set of real-world images and improve accuracy in identifying trailer angles across various conditions without requiring extensive real-world data collection.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If large datasets covering all possible trailer appearances and environmental conditions are used for training, then accuracy in determining trailer angles is improved, but training time and cost increase significantly

Engineering Contradiction:

Improvetrailer angle determination accuracyVSAvoidtraining time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent creates synthetic copies of real-world images through photorealistic rendering software and generative adversarial networks. These synthetic images replicate various trailer appearances and environmental conditions without requiring physical collection of every possible scenario, thus maintaining training accuracy while dramatically reducing data collection time

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent performs preliminary generation of synthetic training data before actual model training. By pre-generating diverse trailer images with known ground truth angles through rendering software, the system prepares comprehensive training datasets in advance, eliminating the need for time-consuming real-world data collection during the training process

Inventive Principle:

Principle #10Preliminary action

2Measurement precision

If large datasets covering all possible trailer appearances and environmental conditions are used for training, then accuracy in determining trailer angles is improved, but training cost increases significantly

Engineering Contradiction:

Improvetrailer angle determination accuracyVSAvoidtraining data volume

Core Design Contradiction:

Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent creates synthetic copies of real-world images through photorealistic rendering software and generative adversarial networks. These synthetic images replicate various trailer appearances and environmental conditions without requiring physical collection of every possible scenario, thus maintaining training accuracy while dramatically reducing data collection time

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent systematically varies parameters in synthetic image generation including trailer types, angles, environmental conditions, and lighting scenarios. By controlling these parameters through software, the system generates diverse training data with known ground truth values without proportionally increasing actual data volume or storage requirements

Inventive Principle:

Principle #35Parameter changes

3Adaptability or versatility

If photorealistic rendering software and generative adversarial networks are used to generate synthetic images, then data diversity is improved, but system complexity increases

Engineering Contradiction:

Improvetraining data diversityVSAvoidsystem complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent introduces photorealistic rendering software and generative adversarial networks as intermediary systems between real-world data and neural network training. These intermediaries translate limited real images into diverse synthetic images with known ground truth, bridging the gap without requiring direct collection of all possible training scenarios

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent employs a universal synthetic data generation system that can handle multiple trailer types, environmental conditions, and viewing angles through a single photorealistic rendering platform. This multi-functional approach replaces the need for separate data collection processes for each scenario, simplifying the overall system architecture

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS20220092356A1Vehicle neural network

Publication Date: 2022.03.24 FORD GLOBAL TECH LLC

AI summary

A system, including a processor and a memory, the memory including instructions to be executed by the processor train a deep neural network based on plurality of real-world images, determine the accuracy of the deep neural network is below a threshold based on identifying one or more physical features by the deep neural network, including one or more object types, in the plurality of real-world images and generate a plurality of synthetic images based on the accuracy of the deep neural network is below a threshold based on identifying the one or more physical features using a photo-realistic image rendering software program and a generative adversarial network. The instructions can include further instructions to retrain the deep neural network based on the plurality of real-world images and the plurality of synthetic images and output the retrained deep neural network.