Sky Segmentation Using Bimodal Intensity and Horizon Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing sky segmentation methods for autonomous vehicles are inefficient, particularly in real-time navigation, as they require significant time investments for dataset creation and algorithm training, and struggle with accurately distinguishing between sky and ground regions in images with only sky or ground present.
Innovation Solution
A method that uses a Canny edge detector to identify strong edges, determines the intensity distribution of pixels, approximates histograms with unimodal and bimodal Gaussian distributions, and employs Otsu's thresholding to divide images into sky and ground regions, allowing for real-time segmentation and horizon detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning semantic segmentation is used to segment sky regions, then segmentation accuracy is improved, but processing time and computational resources increase significantly
Solution Approach 1:
The patent segments the image processing task into distinct stages: edge detection using Canny algorithm, histogram computation, and Gaussian distribution fitting. This segmentation allows each stage to use optimized algorithms appropriate to its specific function, avoiding the overhead of training deep learning models while maintaining segmentation accuracy through the mathematical properties of Gaussian distribution fitting.
Solution Approach 2:
The patent replaces the mechanical training process of deep learning models with a mathematical approach using Gaussian distribution fitting. Instead of training neural networks on large datasets, the system uses closed-form mathematical solutions to fit Gaussian distributions to pixel intensity histograms, achieving comparable accuracy with significantly reduced computational requirements and processing time.
2Productivity
If color or intensity information is used to divide the image into two regions, then processing speed is improved, but accuracy in distinguishing sky and ground regions deteriorates
Solution Approach 1:
The patent transforms the simple color/intensity thresholding approach into a more sophisticated parameter-based method by fitting Gaussian distributions to the intensity histogram. This changes the parameter representation from raw intensity values to distribution parameters (mean, standard deviation, weight), enabling more accurate region distinction while maintaining processing speed through efficient mathematical operations.
Solution Approach 2:
The patent performs preliminary edge detection using the Canny algorithm before applying Gaussian distribution fitting. This preliminary action identifies strong edges that help constrain the segmentation process, improving the accuracy of sky-ground region distinction by providing prior information about potential boundaries before the main segmentation operation.
3Device complexity
If gradient-based methods are used to detect the horizon as the longest horizontal edge, then processing simplicity is improved, but accuracy in complex scenes deteriorates
Solution Approach 1:
The patent introduces Gaussian distribution fitting as an intermediary step between edge detection and horizon identification. Instead of directly finding the longest horizontal edge, the system first fits Gaussian distributions to intensity histograms, uses these distributions to segment the image, and then identifies the horizon. This intermediary mathematical transformation improves accuracy in complex scenes by providing a more robust segmentation basis.
Data Source
AI summary
A method includes identifying a first image. The method also includes determining a distribution based at least partially upon an intensity of each pixel in the first image. The method also includes determining that the distribution is bimodal. The method also includes dividing the first image to produce a second image in response to determining that the distribution is bimodal. The second image includes a plurality of first pixels and a plurality of second pixels. The method also includes determining that a horizon is defined between the plurality of first pixels and the plurality of second pixels.