Autonomous Vehicle Vision for Distant Object Recognition Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image-recognition networks for autonomous vehicles struggle to accurately recognize distant objects due to down-sampling, which reduces pixel size and degrades recognition success rates.
Innovation Solution
A method involving a vehicle's camera to generate a cropped image of a distant region at original resolution, perform first object recognition, and combine it with a down-sampled image's recognition result, using both a first and second object recognition network.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If down-sampling is applied to reduce image resolution for processing, then processing speed and energy consumption are improved, but distant object recognition accuracy deteriorates
Solution Approach 1:
The patent divides the image processing into two segments: a first object recognition network processes down-sampled images for general scene understanding, while a second object recognition network processes cropped high-resolution images of distant regions for accurate distant object detection. This segmentation allows each network to specialize in different resolution requirements, resolving the contradiction between processing speed and distant object recognition accuracy.
Solution Approach 2:
The patent applies different quality levels of image processing to different regions: down-sampled resolution is used for the overall scene (most regions), while original high resolution is preserved for distant regions containing objects of interest. This local quality differentiation ensures that computational resources are concentrated where needed for distant object recognition, maintaining accuracy without sacrificing overall processing efficiency.
2Measurement precision
If original resolution is maintained for distant regions, then distant object recognition accuracy is improved, but computational complexity and processing time increase
Solution Approach 1:
The patent segments the image into processed regions (down-sampled) and unprocessed regions (original resolution), with only the cropped distant regions maintained at full resolution. This segmentation reduces the total computational load compared to processing the entire image at original resolution, while still achieving high distant object recognition accuracy where it matters most.
Solution Approach 2:
The patent applies full-resolution processing only partially to specific distant regions rather than to the entire image. By cropping only the necessary distant regions and processing those at full resolution while down-sampling the rest, the system achieves sufficient recognition accuracy without the excessive computational complexity of processing all pixels at maximum resolution.
Data Source
AI summary
The present disclosure relates to a method and device for a distant object in a vehicle capable of autonomous driving. A method performed by an apparatus of a vehicle may include obtaining, via a camera of the vehicle, an image of an exterior view from the vehicle, generating a cropped image of a distant region in the obtained image, performing first object recognition on the cropped image of the distant region, wherein the cropped image has an original resolution of the obtained image, performing second object recognition on a processed image associated with the obtained image, wherein the processed image has a down-sampled resolution of the obtained image, performing third object recognition by matching a result of the first object recognition with a result of the second object recognition, and controlling, based on a result of the third object recognition, an operation of the vehicle.