Camera Motion Estimation Using Pixel Flow and Feature Points
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional camera motion estimation techniques face challenges in improving estimation accuracy, particularly in scenes with many similar patterns, such as multiple objects like walls, where distinguishing feature points is difficult.
Innovation Solution
A camera motion estimation device that includes a first acquisition unit for acquiring feature point correspondence information and a second acquisition unit for acquiring pixel flow information, using these to estimate camera motion by determining similarity between feature point and pixel moving vectors, thereby enhancing estimation accuracy even in complex scenes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional feature point correspondence techniques are used, then the method is simple to implement, but the estimation accuracy deteriorates in scenes with many similar patterns
Solution Approach 1:
The patent merges feature point correspondence information with pixel flow information to estimate camera motion. By combining these two types of information, the system achieves more accurate motion estimation in complex scenes with many similar patterns, while maintaining reasonable processing complexity through efficient integration methods.
Solution Approach 2:
The patent introduces pixel flow information as an intermediary to bridge the gap between feature points and camera motion estimation. This intermediary provides additional motion cues that help disambiguate feature correspondences in scenes with many similar patterns, improving accuracy without requiring direct complex analysis of all image data.
2Measurement precision
If only feature point correspondence is used, then the processing is computationally efficient, but the accuracy deteriorates when many objects exist in the scene
Solution Approach 1:
The patent applies partial action by selectively using pixel flow information only where it provides additional value - specifically in regions with many similar patterns. This approach improves accuracy in challenging areas while maintaining computational efficiency in simpler regions, achieving a balance between precision and productivity.
Solution Approach 2:
The system applies different processing qualities to different regions of the image. In areas with many similar patterns, it uses the more computationally intensive pixel flow analysis, while in simpler regions, it relies on the more efficient feature point correspondence method, optimizing the trade-off between accuracy and processing efficiency locally.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
According to an arrangement, a camera motion estimation device (20) includes a first acquisition unit (20C), a second acquisition unit (20D), and an estimation unit (20F). The first acquisition unit (20C) is configured to acquire feature point correspondence information indicating correspondence between feature points included in time series images. The second acquisition unit (20D) is configured to acquire pixel flow information indicating a locus of a pixel included in the time series image. The estimation unit (20F) is configured to estimate a motion of a camera (10B) that has captured the time series images, using the pixel flow information and the feature point correspondence information.