Vehicle Camera Image Conversion for Nighttime Object Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous driving systems face limited diversity in learning data, particularly lacking nighttime images, which deteriorates recognition performance during night driving due to the reliance on daytime data acquisition.
Innovation Solution
A vehicle control method that processes image data to filter out light sources and textures, converting daytime images into nighttime images using techniques like unsupervised image-to-image translation networks, ensuring diverse learning data for improved deep learning model performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the autonomous driving system uses deep learning with real-time images as learning data, then the recognition performance is improved, but the diversity of learning data is limited because images can only be acquired when the vehicle is actually driving
Solution Approach 1:
The system performs preliminary image processing by converting daytime images to nighttime images in advance, creating synthetic learning data before actual nighttime driving scenarios occur. This allows the deep learning model to be trained on diverse conditions without requiring the vehicle to physically drive in all scenarios.
Solution Approach 2:
The system creates synthetic copies of daytime images and transforms them into nighttime images through image conversion techniques. These copied and transformed images serve as artificial learning data, expanding the diversity of training data without requiring actual nighttime captures.
2Reliability
If the vehicle collects more diverse learning data by driving in different conditions, then the deep learning performance is improved, but the time and operational constraints increase
Solution Approach 1:
Instead of physically driving the vehicle in different conditions to collect diverse images, the system creates synthetic copies of existing daytime images and converts them to nighttime images. This eliminates the time-consuming process of actual nighttime driving data collection while still providing diverse training data.
Solution Approach 2:
The system changes the temporal parameter of the images by converting daytime images to nighttime images through image processing. This parameter transformation allows the same physical image to represent different temporal conditions, expanding data diversity without additional data collection time.
3Adaptability or versatility
If the system converts all daytime images to nighttime images, then the diversity of learning data is improved, but the processing complexity and computational resources increase
Solution Approach 1:
The system extracts and removes specific elements (light sources, textures) from daytime images before converting them to nighttime images. This selective extraction simplifies the conversion process by focusing only on relevant features, reducing overall processing complexity while maintaining data diversity.
Solution Approach 2:
The system performs preliminary filtering and element removal from images before the conversion process. This preliminary action simplifies subsequent conversion operations and reduces computational complexity by preprocessing images in a structured manner.
Data Source
AI summary
An embodiment vehicle includes a camera and a controller including a processor configured to process image data acquired from the camera, wherein the controller is configured to determine whether a light source or a texture is present in a first image acquired by processing the image data, perform filtering of the first image based on the light source or the texture being present, convert the first image into a second image, and store the first image and the second image as learning data for object recognition.