计算机科学 ›› 2017, Vol. 44 ›› Issue (3): 36-37.doi: 10.11896/j.issn.1002-137X.2017.03.009

• 2015全国高性能计算学术年会 • 上一篇    下一篇

基于Pthreads的车辆图像兴趣区域提取并行算法研究

周艺华,王文东,陈宏彩,王婷,张常有   

  1. 北京工业大学计算机学院 北京100124;可信计算北京市重点实验室 北京100124,北京工业大学计算机学院 北京100124;中国科学院软件研究所并行软件与计算科学实验室 北京100190,河北省应用数学研究所 石家庄050081,中国科学院软件研究所并行软件与计算科学实验室 北京100190,中国科学院软件研究所并行软件与计算科学实验室 北京100190
  • 出版日期:2018-11-13 发布日期:2018-11-13
  • 基金资助:
    本文受国家自然科学基金项目(61379048),中国科学院联盟院地合作项目:面向企业创新计算的高性能云服务平台和智能交通中的关键技术研究资助

Research on Parallel Algorithm of Vehicle Image Interested Region Based on Pthreads

ZHOU Yi-hua, WANG Wen-dong, CHEN Hong-cai, WANG Ting and ZHANG Chang-you   

  • Online:2018-11-13 Published:2018-11-13

摘要: 为了提高公安机关查找犯罪车辆的效率,提高车辆识别的效率很必要。据统计,提取兴趣区域(Region Of Interest,ROI)约占车型识别过程的60%,因此如何加速提取ROI过程尤其重要。首先,通过数据划分方法实现基本并行算法;然后,经过实验分析,在基本并行算法的基础上,精心设计预处理过程的分解方案,设置多队列缓冲区,减少共用缓冲区的线程数量和每个缓冲区互斥锁锁定的次数。实验证明,所提算法在双CPU 12核(支持超线程到24线程)的服务器上运行,相对于串行算法,实现了13.1x的加速比。

关键词: 车型识别,兴趣区域,并行化,Pthreads,多核

Abstract: In order to improve the efficiency of searching criminal vehicles for the public security bureau,improving the efficiency of vehicle identification is very necessary.According to statistics,extraction of region interest area (ROI) is about 60% of the entire vehicle identification process.How to speed up the phase of extracting ROI is especially important.First,the basic parallel algorithm was realized by data partitioning method.Then,through the experimental analysis,the pre-processing decomposition scheme was carefully designed based on basic parallel algorithm.In our scheme,we set up a multi-queue buffer,reducing the number of threads sharing a cache and the times of lock each cache.Experiments show that the algorithm running on server with dual CPU12 core (support for hyper threading to 24 threads) achieved 13.1x speedup ratio compared with the serial algorithm.

Key words: Vehicle identification,Region of interest,Parallelization,Pthreads,Multi core

[1] FREJLICHOWSKI D,GRZEGORZEWICZ K.An Approach to Automatic Detection and Extraction of Regions of Interest in Still Images[M]∥Image Processing and Communications Challenges 4.Spring Berlin Heidelberg,2013:3-10.
[2] HIO C,BERMINGHAM L,CAI G,et al.A Hybrid Grid-based Method for Mining Arbitrary Regions-of-Interest from Trajectories[C]∥The Workshop on Machine Learning for Sensory Data Analysis.2013 .
[3] EHTESHAMI N S M,TABANDEH M,FATEMIZADEH E.A new ROI extraction method for FKP images using global intensity[C]∥2012 Sixth International Symposium on Telecommunications (IST).IEEE,2012:1147-1150.
[4] SAIFULLAH A,LI J,AGRAWAL K,et al.Multi-core real-timescheduling for generalized parallel task models[J].Real-Time Systems,2013,49(4):217-226.
[5] LIANG H,LIU R,GUO W.Performance of the Buffer Queue With Priority For Dynamic Spectrum Access[C]∥2010 International Conference on Advanced Intelligence and Awarenss Internet (AIAI 2010).2010:109-112.
[6] BERGAN T,CEZE L,DAN G.Input-Covering Schedules forMultithreaded Programs[J].ACM Sigplan Notices,2013,48(10):677-692.
[7] CHEN G,STENSTROM P.Critical lock analysis:Diagnosingcritical section bottlenecks in multithreaded applications[C]∥ Proceedings of the 2012 International Conference for High Performance Computing,Networking,Storage and Analysis.IEEE Computer Society,2012:1-11.
[8] DICE D,MARATHE V J,SHAVIT N.Lock cohorting:a gene-ral technique for designing NUMA locks[J].ACM Sigplan Notices,2012,47(8):247-256.

No related articles found!
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
No Suggested Reading articles found!