Vanishing Point Extraction Using Temporal Feature Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional vanishing point extraction methods fail to extract vanishing points from images when lanes are not detected, which is crucial for autonomous driving applications, especially in environments with no lane markings or unclear lanes.
Innovation Solution
A method and device that extract vanishing points from a current image by utilizing information from a previous image, including identifying a straight line and sample points within the previous image, and matching these points in subsequent images using template matching and correlation analysis, even if lanes are not detected in the current image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If lane detection method is used to extract vanishing point, then vanishing point can be extracted from images with clear lane markings, but vanishing point cannot be extracted when lanes are not detected or unclear
Solution Approach 1:
The patent applies preliminary action by extracting feature points and establishing correspondence relationships between consecutive images before vanishing point extraction. By pre-processing images to identify corners, edges, and other geometric features, and by establishing matching relationships between these features across time, the system prepares necessary data in advance that enables vanishing point calculation even when lane markings are absent or unclear in the current image.
Solution Approach 2:
The patent uses feature points and their correspondence relationships as intermediaries between images. Instead of directly detecting vanishing points from lane markings in each image, the system extracts intermediate feature points (corners, edges, contours) and establishes their temporal correspondence. These intermediate features serve as mediators that bridge consecutive images, allowing the vanishing point to be inferred from the geometric relationships of matched features rather than requiring direct lane detection.
2Ease of manufacture
If conventional lane detection method is used, then extraction process is simple when lanes are present, but extraction fails completely when lanes are absent
Solution Approach 1:
The patent segments the vanishing point extraction process into multiple independent stages: feature point extraction, correspondence establishment, and vanishing point calculation. By dividing the overall task into separate modules that can process different types of visual information (lane markings, general geometric features, temporal relationships), the system maintains simplicity in each individual stage while achieving robustness through the combination of these segmented processing steps.
3Measurement precision
If template matching is used to find corresponding points between images, then accurate matching can be achieved, but computational complexity increases
Solution Approach 1:
The patent applies local quality by performing template matching only in localized regions around detected feature points rather than across the entire image. By identifying key feature points first (such as corners or distinctive geometric structures) and then applying template matching only in their vicinity, the system achieves accurate local correspondence while avoiding the computational burden of global template matching across the whole image.
Data Source
AI summary
Vanishing point extraction includes obtaining a straight line including a vanishing point of a first image; obtaining a plurality of sample points in the first image based on processing the first image according to an object included in the first image and the straight line including the vanishing point of a first image, such that the plurality of sample points are determined as pixels in the first image having coordinates that overlap with coordinates of pixels of both the straight line and the object included in the first image; obtaining at least one matching point, in a second image, that corresponds to at least one sample point of the plurality of sample points in the first image the second image generated subsequently to the first image being generated; and obtaining a vanishing point of the second image based on the at least one matching point of the second image.