Stereo Image Lane Determination via Segmented Disparity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for estimating depth information from stereo images, such as those used in advanced driver assistant systems (ADAS) and autonomous navigation, face challenges in calculating disparity across the entire road surface, leading to long calculation times and difficulties in predicting lanes, especially when road segments are obscured by obstacles.
Innovation Solution
A method that extracts straight-line segments from stereo images, selects a primary lane by clustering these segments, predicts additional lane candidates through extrapolation and confidence value calculations based on gradient direction and curvature, and updates their positions using disparity correction, thereby enhancing lane detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If disparity calculation is performed across the entire road surface, then depth information accuracy is improved, but calculation time increases significantly
Solution Approach 1:
The patent divides the road surface into multiple regions of interest (ROIs) based on lane line segments detected in the image. Instead of calculating disparity across the entire image, the system selectively calculates disparity only within these identified ROI areas. This segmentation approach maintains depth information accuracy for lane detection while significantly reducing the computational burden and calculation time.
2Productivity
If sparse depth estimation is used to reduce calculation time, then processing speed is improved, but additional calculations are required to distinguish roads and lanes
Solution Approach 1:
The patent performs preliminary detection of lane line segments and identification of regions of interest before conducting disparity calculations. By pre-processing the image to identify where lane information is likely to be located, the system prepares the necessary spatial context in advance. This preliminary action enables sparse disparity calculation to be performed efficiently without requiring additional complex post-processing steps to distinguish roads and lanes.
3Area of stationary object
If traditional disparity calculation methods are used, then complete road surface coverage is achieved, but lane prediction becomes difficult when road portions are obscured by obstacles
Solution Approach 1:
The patent introduces lane line segment detection and region of interest identification as intermediary steps between image acquisition and disparity calculation. These intermediaries help the system focus on relevant areas (lane markings) even when parts of the road are obscured. By using these intermediary detection mechanisms, the system can predict lane positions reliably based on visible line segments and their geometric relationships, rather than relying on complete road surface visibility.
Data Source
AI summary
Disclosed is a method and apparatus for determining a lane, the method including extracting plural straight-line segments from a captured stereo image including a first image and a second image of a driving road, selecting a first lane from the stereo image based on line segments of the plural straight-line segments, corresponding to lanes of the driving road, predicting a second lane candidate, based on the first lane, including at least a portion of a lane area excluding the line segments in the stereo image, and determining a second lane by updating a position of the predicted second lane candidate based on a confidence value of the predicted second lane candidate.