GPU Extreme Point Detection via DOG Layer Comparison
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for detecting features in images, such as SIFT, are computationally intensive and power-hungry, posing challenges for energy-efficient implementation on graphics processing units (GPUs) with limited memory and computing capacity, especially in mobile devices.
Innovation Solution
A method involving a graphics processing unit (GPU) that processes a difference of Gaussian (DOG) layer to detect extreme points by comparing values across layers, using a shader core and buffer to store and localize key points, optimizing computations through fixed-point arithmetic and parallel processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional feature detection methods (SIFT) are implemented on GPU, then feature detection capability is improved, but energy consumption increases
Solution Approach 1:
The patent segments the feature detection process into distinct stages: DOG layer generation, extreme point detection, and key point localization. Each stage is processed independently on the GPU, allowing for optimized resource allocation and energy management at each segment rather than processing the entire SIFT pipeline sequentially
Solution Approach 2:
The patent implements partial action by detecting only extreme points in the DOG layers rather than performing complete SIFT feature extraction. This partial processing approach maintains essential feature detection capability while significantly reducing computational overhead and energy consumption on mobile GPUs
2Speed
If SIFT features are extracted in real-time, then processing speed is improved, but computation resources increase
Solution Approach 1:
The patent extracts only the essential extreme point information from DOG layers, removing unnecessary computational steps of full SIFT processing. By taking out only the critical detection phase and performing key point localization separately, the system achieves real-time processing with reduced computation resource requirements
Solution Approach 2:
The patent performs preliminary action by pre-computing DOG layers and identifying extreme points before final key point localization. This preliminary processing organizes data in advance, enabling faster real-time processing during actual feature extraction while distributing computation resources more efficiently
3Productivity
If parallelization of SIFT is implemented, then productivity is improved, but load imbalance occurs
Solution Approach 1:
The patent segments the image processing into independent tile-based regions, allowing parallel processing across multiple GPU cores. Each tile is processed independently for DOG layer generation and extreme point detection, achieving high productivity while avoiding load imbalance through uniform work distribution across processing units
Data Source
AI summary
Provide are a methods and devices for processing graphics data in a graphics processing unit (GPU). The method of processing graphics data includes receiving, at a processor, a difference of Gaussian (DOG) layer of an image, detecting, from the received DOG layer, a candidate DOG layer of the image as an intermediate layer, detecting at least one extreme point by comparing values of the candidate DOG layer with values of a previous DOG layer and a next DOG layer, and storing the at least one extreme point in a buffer.