Feature Point Matching With Direction-Limited Search in vSLAM
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods for feature point matching in SfM and vSLAM require significant computational complexity, leading to prolonged processing times, and previous methods to reduce this complexity have limitations that can result in incomplete or inefficient processing.
Innovation Solution
An image processing apparatus that limits the search range for feature point matching by switching the reference frame based on the advancing direction of a movable apparatus, using a camera system with multiple cameras capturing images in directions parallel to or opposite to the apparatus's direction, and employing cell division and a pre-determined search range map to reduce computational complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If the search range is limited based on self-position estimation information, then computational complexity is decreased, but processing time increases due to additional processing steps
Solution Approach 1:
The patent pre-divides the image into multiple cells and pre-determines search range maps for each cell based on the camera's optical axis direction. This preliminary organization allows the system to directly reference pre-computed search ranges without performing complex real-time calculations, thus reducing computational complexity while maintaining real-time processing capability.
Solution Approach 2:
The patent divides the image into multiple cells and creates separate search range maps for each cell. This segmentation allows the system to process only relevant portions of the image rather than the entire image, significantly reducing the computational complexity of feature point matching while enabling faster processing through parallel cell evaluation.
2Device complexity
If the search range is set from the vanishing point to the edge of the image, then the operational complexity for detecting a movable apparatus is reduced, but the processing amount cannot be sufficiently decreased due to the position of the vanishing point
Solution Approach 1:
The patent assigns different search range characteristics to different cells based on their local properties. Each cell has its own search range map that is optimized for its specific position and orientation relative to the camera's optical axis. This local optimization allows the system to reduce processing amounts more effectively than a uniform search range approach, as each cell's search range is tailored to its specific geometric context.
Solution Approach 2:
The patent changes the parameter of search range definition from a fixed vanishing-point-based approach to a cell-specific approach based on the camera's optical axis direction. By adjusting the search range parameters according to each cell's relationship to the optical axis, the system achieves more significant reductions in processing amount while maintaining detection accuracy.
3Measurement precision
If corresponding points are searched for from all of the input images, then matching accuracy is maintained, but a vast computational complexity becomes necessary and processing time becomes long
Solution Approach 1:
The patent pre-organizes the image space into cells with predetermined search range maps that encode the geometric relationships relative to the camera's optical axis. This preliminary structuring enables the system to maintain matching accuracy by ensuring that all relevant corresponding points are included in the search, while simultaneously reducing computational complexity by limiting the search to pre-defined ranges rather than examining all possible point combinations.
Solution Approach 2:
By dividing the image into cells and creating separate search range maps for each, the patent enables targeted searching within localized regions. This segmentation maintains matching accuracy by ensuring that corresponding points within each cell are properly identified, while reducing overall computational complexity by preventing unnecessary searches across the entire image space.
Data Source
Figure 1
Figure 2A~2B
Figure 3
AI summary
An image processing apparatus (1000) includes image capturing position movement direction limiting means for selecting, from plural time series images acquired from an image capturing means (10) disposed on a movable apparatus (1000) for capturing surrounding images, first and second images captured when the movable apparatus was at first and second positions, respectively, and controlling a movement direction of an image capturing position to the first position and the second position; feature point matching means (230) for selecting a reference image and a search subject image from the first image and the second image, and performing feature point association; and reference image switching means (220) for switching the reference image based on information showing a movement direction of the movable apparatus. The feature point matching means searches for each feature point in the search subject image within a search range limited based on the movement direction of the movable apparatus.