3D Scale Determination Using Localized Ground Plane Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating 3D information from images captured by a vehicle-mounted camera face challenges in accurately determining the scale due to the mis-identification of the ground plane, leading to incorrect distances and potential vehicle collisions in applications like free parking space detection.
Innovation Solution
A method that identifies candidate points associated with a reference feature within a specific portion of the camera's field of view, using techniques like RANSAC to refine the ground plane identification and determine the scale based on the distance between the camera and the reference feature, thereby reducing errors in scale determination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the ground plane is identified by selecting candidate points from the entire image based on maximum feature density along the Y axis, then the method is simple and computationally efficient, but the ground plane may be incorrectly identified leading to inaccurate scale determination
Solution Approach 1:
The patent applies local quality by restricting the search for ground plane candidate points to a specific region of interest (lower portion of the image) rather than the entire image. This localized approach improves measurement precision by focusing computational resources on the area most likely to contain the actual ground plane, while still maintaining reasonable computational efficiency through targeted processing.
Solution Approach 2:
The patent segments the image processing task by dividing candidate point selection into different spatial regions. By separating the ground plane search from other feature detection tasks and confining it to the lower portion of the image, the method reduces the risk of selecting false ground plane candidates from other horizontal surfaces while maintaining computational efficiency.
2Reliability
If candidate points are selected from the entire image to identify the ground plane, then more data is available for analysis, but the likelihood of false identification increases due to multiple horizontal surfaces at different heights
Solution Approach 1:
The patent improves reliability by applying local quality constraints to the ground plane identification process. By limiting candidate point selection to the lower portion of the image where the actual ground plane is most likely to be located, the method reduces false identifications from other horizontal surfaces while maintaining a relatively simple implementation approach.
3Adaptability or versatility
If the scale is determined using camera height ratio based on ground plane identification, then the scale can be calculated without prior knowledge of real world feature sizes, but incorrect ground plane identification leads to wrong scale values
Solution Approach 1:
The patent resolves this contradiction by applying local quality to the ground plane identification process. By restricting candidate point selection to the lower portion of the image, the method maintains its adaptability and versatility in determining scale without prior knowledge of real world feature sizes, while simultaneously improving scale measurement accuracy by reducing false ground plane identifications.
Data Source
AI summary
This disclosure relates to a method in which points common to two or more of the images that appear to represent the same real world features are identified, and changes in the location of the points between respective images are used to deduce the motion of the camera and to find the position of the real world features in three dimensional space. In order to determine the scale of the three dimensional information, the position of a reference feature, whose actual distance from the camera is known, is found from the images. The reference feature is found by considering only candidate points selected from candidate points falling within a portion of the image corresponding to a part of the field of view of the camera. The scale is determined from the distance between the camera and the reference feature in the image and in real life.