Feature Detection Using Segmented FAST and SIFT Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Computer vision algorithms, such as SIFT, are resource-intensive and complex, making them unsuitable for mobile devices, while faster but less accurate detectors like FAST often result in false identifications due to noise and clutter in images, limiting their effectiveness in object recognition on portable devices.
Innovation Solution
Implementing a method that uses a computationally inexpensive feature detector like FAST to segment images and isolate regions of interest, followed by a more accurate detector like SIFT within those regions, to improve object recognition speed and accuracy without overwhelming mobile device resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a computationally intensive feature detector like SIFT is applied to the entire image, then object recognition accuracy is improved, but processing time increases and computational resources are overwhelmed
Solution Approach 1:
The patent divides the image processing task into two stages: first applying a fast feature detector (FAST) to the entire image to identify candidate regions, then applying the more accurate but computationally intensive detector (SIFT) only to those candidate regions. This segmentation of the processing domain reduces the total computational load while maintaining recognition accuracy for objects of interest.
Solution Approach 2:
The patent applies different levels of detection quality to different regions of the image. Candidate regions identified by the fast detector receive high-quality SIFT analysis, while non-candidate regions are processed only by the fast detector. This local differentiation of processing quality optimizes the balance between accuracy and computational cost.
2Productivity
If a fast feature detector like FAST is applied to the entire image, then processing speed is improved, but object recognition accuracy deteriorates due to false positives from noise and clutter
Solution Approach 1:
The patent segments the image into candidate regions and non-candidate regions based on fast detector results. By then applying the more accurate SIFT detector specifically to candidate regions, the system eliminates false positives from noise and clutter while maintaining high processing speed through the initial fast detection phase.
Solution Approach 2:
The patent performs preliminary filtering using the fast detector to identify candidate regions before applying the more accurate SIFT detector. This preliminary action reduces the search space and eliminates obvious false positives early, allowing the system to maintain speed while improving accuracy in the subsequent detailed analysis phase.
3Measurement precision
If a computationally intensive feature detector is applied to the entire image, then detection accuracy is improved, but device complexity and resource requirements increase
Solution Approach 1:
The patent segments the computational task by applying different detectors to different portions of the image. The fast detector handles the entire image with low computational complexity, while the intensive SIFT detector is applied only to small candidate regions. This segmentation dramatically reduces overall device complexity and resource requirements while maintaining high detection accuracy where needed.
Solution Approach 2:
The patent applies high-quality SIFT detection only locally to candidate regions rather than globally to the entire image. This local application of high-quality detection maintains accuracy for objects of interest while significantly reducing the total computational complexity and resource requirements of the system.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
In a particular embodiment, a method includes applying a first feature detector to a portion of an image to detect a first set of features. The first set of features is used to locate a region of interest, and a boundary corresponding to the region of interest is determined. The method also includes displaying the boundary at a display. In response to receiving user input to accept the displayed boundary, a second feature detector is applied to an area of the image encapsulated by the boundary.