Comparison of Image Processing Algorithms on Micro-Array Architectures and GPGPU Platforms | Consortium for Embedded Systems | SIU

Abstract: The proposed project is a study analyzing the performance of image processing algorithms on General-Propose computing on Graphics Processing Units (GPGPU) architecture in comparison to a micro-core array embedded processor. The project will create a set of metrics which will measure speed, power consumption, and cost. The metrics will provide an exact analysis of the trade-offs between the different architectures. Furthermore, the study will determine feasibility for processing real-time video information of high resolution video streams. Additionally, an investigation into the performance of several image processing algorithms such as a noise reduce, will be performed. A core objective will be to select several image processing algorithms and optimize them for the GPGPU environment. Analysis of the performance gains and trade-offs will be included in the final survey.

Problem:

• Currently no comprehensive survey comparing existing embedded systems processors to emerging mobile GPGPU platforms

Rationale / Approach:

• Focus shall be on implementation of image processing algorithms for in-house micro-core array processor and on GPGPU platforms with at least one mobile GPGPU platform.
• Optimizations shall be made to the set of image processing algorithms

Novelty:

• First Study to compare mobile versions of GPGPU the platform to current embedded systems
• Optimizations to existing Image Processing Algorithms will be made to better parallelize them

Potential Member Company Benefits:

• Member companies will gain additional insight into hardware platform trade-offs for future designs
• Gives better insight into implementation of image processing algorithms in embedded systems at a time when vision sensors are becoming commonplace.

Deliverables for the proposed year:

• Survey of GPGPU Platforms
• Comparison Study of the embedded system and GPGPU Platforms
• GPGPU Optimized Image Processing Algorithms

Milestones for the proposed year:

• Select GPGPU Platform (8/31/2014)
• Select a set of image processing algorithms (9/30/2014)
• Implement and optimize image processing algorithms on GPGPU platform (12/20/2014)
• Port algorithms to a multi-core embedded system (3/31/2015)
• Perform an Analysis on the performance difference between the platforms (7/31/2015)

Progress to Date:

• Acquired infrastructure for CUDA projects
• Previous experience in implementing parallel image processing algorithms