Object Detection Neural Network Fusing RGB and Infrared Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection methods, such as those using RGB images, face accuracy issues in low-light environments due to limited information capture, leading to insufficient object detection results.
Innovation Solution
The integration of multiple image types, like RGB and infrared images, is used to create a fused image for training an object detection neural network, enhancing detection accuracy across various lighting conditions by combining features from both image types.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single source image (RGB image) is used for object detection, then the device complexity is low, but the measurement precision deteriorates in dark environments
Solution Approach 1:
The patent combines multiple types of images (RGB image and infrared image) into a fused image for object detection. The image generation unit creates an infrared image from the RGB image, and the image fusion unit merges them to produce a fused image that contains both color information and thermal information, thereby improving detection accuracy in dark environments while managing device complexity through software-based processing.
2Measurement precision
If multiple types of images (RGB and infrared) are fused, then the object detection accuracy improves, but the loss of time increases due to additional processing
Solution Approach 1:
The image generation unit pre-generates an infrared image from the RGB image by converting color information to grayscale and applying gamma correction. This preliminary processing allows the fusion unit to efficiently combine the pre-processed infrared image with the original RGB image, reducing the overall processing time compared to generating all processed images in sequence during detection.
3Adaptability or versatility
If an infrared image is generated from an RGB image, then the adaptability to low-light conditions improves, but the loss of information occurs during color to grayscale conversion
Solution Approach 1:
The image fusion unit merges the RGB image (which contains color information) with the infrared image (which contains thermal information suitable for low-light detection). This fusion preserves the color information from the original RGB image while adding the lighting-condition-robust thermal information, thereby achieving adaptability to low-light conditions without permanently losing color information.
Data Source
AI summary
An image processing device including a storage unit configured to store an object detection neural network trained using a first K-channel image generated from a first M-channel image and a first N-channel image generated from the first M-channel image, a reception unit configured to receive, from a sensor, a second M-channel image and a second N-channel image that include an identical subject, and an image analysis unit configured to generate, using the object detection neural network trained using the first K-channel image, object detection result information with respect to a second K-channel image generated from the second M-channel image and the second N-channel image, and output the object detection result information.