Vehicle Image Augmentation for Trailer Orientation Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in determining the orientation of unfamiliar trailer types due to limitations in existing deep neural networks, which struggle with varying environmental conditions and trailer configurations, leading to failures in determining trailer angles.
Innovation Solution
A method involving a few-shot image translator based on a generative adversarial network is used to generate modified images of trailer types, allowing for re-training of the deep neural network to improve its ability to determine trailer angles in diverse conditions, including daytime, nighttime, rain, snow, and different sunlight directions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a deep neural network is trained on a limited set of trailer types and environmental conditions, then the training process is efficient and quick, but the network fails to accurately determine orientation of unfamiliar trailer types in diverse conditions
Solution Approach 1:
The system performs preliminary action by generating synthetic training images that anticipate diverse trailer types and environmental conditions before actual deployment. The few-shot image translator creates modified images of unfamiliar trailers by translating existing training images, preparing the neural network in advance for various scenarios it may encounter in the wild, thus improving adaptability without requiring extensive real-world data collection for every possible scenario
Solution Approach 2:
The system uses copying by creating synthetic copies of existing trailer images through the few-shot image translator. Instead of requiring numerous real-world images of every trailer type, the system generates modified copies of available training images that simulate different trailer configurations and environmental conditions, allowing the network to learn from these synthesized copies and improve its ability to recognize unfamiliar trailers
2Measurement precision
If more real-world images of diverse trailer types are collected for training, then the deep neural network's accuracy in determining trailer orientation improves, but the data collection time and computational resources increase significantly
Solution Approach 1:
The system creates synthetic copies of existing training images using the few-shot image translator to generate diverse trailer representations without requiring physical collection of numerous real-world images. This copying approach allows the neural network to be trained on a wide variety of trailer types and conditions while avoiding the time-consuming process of gathering and annotating extensive real-world image datasets
Solution Approach 2:
The system applies parameter changes by modifying image parameters through the few-shot image translator to generate diverse training samples. By transforming existing images with different transformations and modifications, the system creates varied training data that improves measurement precision for trailer orientation determination without requiring proportional increases in data collection time
Data Source
AI summary
A system, including a processor and a memory, the memory including instructions to be executed by the processor to receive one or more images from a vehicle, wherein a first deep neural network included in a computer in the vehicle has failed to determine an orientation of a first object in the one or more images. The instructions can include further instructions to generate a plurality of modified images with a few-shot image translator, wherein the modified images each include a modified object based on the first object. The instructions can include further instructions to re-train the deep neural network to determine the orientation of the first object based on the plurality of modified images and download the re-trained deep neural network to the vehicle.