Energy and Execution Time Comparison of Optical Flow Algorithms on SIMD and GPU Architectures

Abstract

This article presents and compares optimized implementations of two optical flow algorithms on several target boards comprising multi-core SIMD processors and GPUs. The two algorithms are Horn-Schunck (HS) and TV-L1, and have been chosen because they are both well-known, and because of their different computational complexity and accuracy. For both algorithms, we have made parallel optimized SIMD implementations, while HS has also been implemented on GPUs. For each algorithm, the comparison between the different versions and target boards is carried out in a two-dimensional fashion: in terms of computing speed-in order to achieve real-time computation-and in terms of energy consumption since we target embedded systems. The results show that for HS, the GPUs are the most efficient in both dimensions, able to process in realtime performances (25 frames per second) up to 8Mpix images for 0.35J per image, against 1.8Mpix images for 0.24J per image on CPU. The results also highlight the impact of optimizations on TV-L1: far slower than HS without optimization, it can almost match its performance after optimization on CPU, and can achieve real-time performances with 0.25J for 1.4Mpix images. We hope these results will help developers design optical flow embedded systems.

An Efficiency-Driven Approach For Real-Time Optical Flow Processing On Parallel Hardware

“…Speeding-up this convergence process is thus required for real-time. To tackle this issue, many recent works on optical flow leverage the massive computing power of GPUs [4,5,6]. Originally reserved for computer graphics and 3D-scenes rendering, they now play a prominent role in image processing.…”

FPGA Acceleration of the Horn and Schunck Hierarchical Algorithm

“…By taking all these into account, depending on the number of cores (II) used and the number of iterations, the total time T for the multi-scale algorithm calculation of the image, without the down-sampling (which takes less than 15% of the total time), can be estimated by (3).…”

A New Real-Time Embedded Video Denoising Algorithm

“…Optical flow algorithms are also time consuming but are sensitive to code transformations and can be highly accelerated [18]- [22]. We thus decided to use an optical flow algorithm in our real-time denoising chain.…”

“…The TV-L1 optical flow algorithm is the main step to optimize as it takes more than 80% of the total time. The key transformations of TV-L1 are presented in [18] and result in a processing 5× faster and 6× less power consuming on an embedded ARM Cortex A57 architecture.…”

“…As a consequence, the optical flow estimation now takes almost 98% of the total time of the Fast version, while filtering was the most time consuming part of the processing chain without the optimizations. Since the optical flow estimation is, in the end, the most consuming step, we produced a more detailed study in [18]. In this previous work the impact of each optimisation is discussed.…”