#95: Multicore energy efficient flux tensor for video analysis

K. Palaniappan, I. Ersoy, G. Seetharaman, S. Davis, R. Rao, and R. Linderman

IEEE Workshop on Energy Efficient High-Performance Computing (EEHiPC), 2010

parallelization, wami, tracking, fmv, motion, dod

PlainText, Bibtex, PDF, URL, Google Scholar


The flux tensor motion flow algorithm is a versatile computer vision technique for robustly detecting moving objects in clut- tered scenes. The flux tensor calculation has a high compu- tational workload consisting of 3-D spatiotemporal filtering operations combined with 3-D weighted integration opera- tions for estimating local averages of the flux tensor matrix trace. In order to achieve efficient real-time processing of high bandwidth video streams a data parallel multicore al- gorithm was developed for the Cell Broadband Engine (BE) processor and evaluated in terms of the energy to computa- tion efficiency compared to a fast sequential CPU implemen- tation. Our multicore implementation is 12 to 40 times faster than the sequential version for HD video using a single PS-3 Cell/B.E. processor and is faster than realtime for a range of filter configurations and video frame sizes. We report on the power efficiency measured in terms of performance per watt for the Cell/B.E. implementation which is 50 to 160 times better than the sequential version for HD video depending on the filter size. The results suggest an additional strategy to trade off output image quality or nominal change in accuracy of detection for improved energy efficiency in suitable envi- ronments.