CUDA implementation of the pixel based adaptive segmentation algorithm

The operation of foreground background subtraction, which has a crucial role on video processing, needs to be processed in real time. In this paper, we present an implement of the pixel based adaptive segmentation (PBAS) algorithm on CUDA parallel programming platform to process high resolution videos in real time. The performance of the implementation is enhanced by using Nsight profiler metrics such as consumed time for functions, usage of GPU compute and memory, occupancy, divergence. Experimental evaluation on benchmark videos presents the performance of the implementation; On a Tesla K20 graphic processing unit, 646 fps is achieved for a video with 320×240 resolution when memory transfer to and from the GPU is included. As a result, about 17 time speedup is achieved with regards to single thread version.

[1]  Z. Zivkovic.  Improved adaptive Gaussian mixture model for background subtraction , 2004, ICPR 2004.

[2]  David Schreiber,et al.  GPU-based non-parametric background subtraction for a practical surveillance system , 2009, 2009 IEEE 12th International Conference on Computer Vision Workshops, ICCV Workshops.

[3]  Hoai Bac Le,et al.  GPU Implementation of Extended Gaussian Mixture Model for Background Subtraction , 2010, 2010 IEEE RIVF International Conference on Computing & Communication Technologies, Research, Innovation, and Vision for the Future (RIVF).

[4]  Fatih Murat Porikli,et al.  Changedetection.net: A new change detection benchmark dataset , 2012, 2012 IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops.

[5]  Gerhard Rigoll,et al.  Background segmentation with feedback: The Pixel-Based Adaptive Segmenter , 2012, 2012 IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops.

[6]  Tomasz Kryjak,et al.  Implementation of Advanced Foreground Segmentation Algorithms GMM, ViBE and PBAS in FPGA and GPU - A Comparison , 2014, ICCVG.