Increasing the Efficiency of GPU-Based HOG Algorithms Through Tile-Images

详细信息    查看全文

• 关键词：Image ; Composed image ; gpgpu ; High performance computing ; Histogram of oriented gradients ; HOG ; opencl ; Cuda
• 刊名：Lecture Notes in Computer Science
• 出版年：2016
• 出版时间：2016
• 年：2016
• 卷：9621
• 期：1
• 页码：708-720
• 全文大小：615 KB
• 参考文献：1.Azmat, S., Wills, L., Wills, S.: Accelerating adaptive background modeling on low-power integrated gpus. In: 2012 41st International Conference on Parallel Processing Workshops (ICPPW), pp. 568–573. IEEE (2012)
  2.Dalal, N., Triggs, B.: Histograms of oriented gradients for human detection. CVPR 1, 886–893 (2005)
  3.Erdos, P., Graham, R.: On packing squares with equal squares. J. Comb. Theor. Ser. A 19(1), 119–123 (1975). http://​www.​sciencedirect.​com/​science/​article/​pii/​0097316575900990​ MathSciNet CrossRef MATH
  4.Gibbons, P.B.: A more practical pram model. In: Proceedings of the First Annual ACM Symposium on Parallel Algorithms and Architectures, pp. 158–168. ACM (1989)
  5.Hirabayashi, M., Kato, S., Edahiro, M., Takeda, K., Kawano, T., Mita, S.: Gpu implementations of object detection using hog features and deformable models. In: 2013 IEEE 1st International Conference on Cyber-Physical Systems, Networks, and Applications (CPSNA), pp. 106–111. IEEE (2013)
  6.Hommel, S., Malysiak, D., Grimm, M., Handmann, U.: Apfel - fast multi camera people tracking at airports, based on decentralized video indexing. Sci.2 - Saf. Secur. 2, 48–55 (2014)
  7.Hommel, S., Malysiak, D., Handmann, U.: Model of human clothes based on saliency maps. In: 2013 IEEE 14th International Symposium on Computational Intelligence and Informatics (CINTI), pp. 551–556, November 2013
  8.Kachouane, M., Sahki, S., Lakrouf, M., Ouadah, N.: Hog based fast human detection. In: 2012 24th International Conference on Microelectronics (ICM), pp. 1–4. IEEE (2012)
  9.Lodi, A., Martello, S., Monaci, M.: Two-dimensional packing problems: A survey. Eur. J. Oper. Res. 141(2), 241–252 (2002). http://​www.​sciencedirect.​com/​science/​article/​pii/​S037722170200123​6 MathSciNet CrossRef MATH
  10.Malysiak, D., Handmann, U.: An efficient framework for distributed computing in heterogeneous beowulf clusters and cluster-management. In: 2014 IEEE 15th International Symposium on Computational Intelligence and Informatics (CINTI), November 2014
  11.Prisacariu, V., Reid, I.: fasthog - a real-time gpu implementation of hog. Technical Report 2310/09. Department of Engineering Science, Oxford University
  12.Sudowe, P., Leibe, B.: Efficient use of geometric constraints for sliding-window object detection in video. In: Crowley, J.L., Draper, B.A., Thonnat, M. (eds.) ICVS 2011. LNCS, vol. 6962, pp. 11–20. Springer, Heidelberg (2011)CrossRef
  13.Yudanov, D., Shaaban, M., Melton, R., Reznik, L.: Gpu-based simulation of spiking neural networks with real-time performance & high accuracy. In: The 2010 International Joint Conference on Neural Networks (IJCNN), pp. 1–8. IEEE (2010)
  
• 作者单位：Darius Malysiak (17)
  Markus Markard (17)
  
  17. Computer Science Institute, Hochschule Ruhr West, Mülheim, Germany
  
• 丛书名：Intelligent Information and Database Systems
• ISBN：978-3-662-49381-6
• 刊物类别：Computer Science
• 刊物主题：Artificial Intelligence and Robotics
  Computer Communication Networks
  Software Engineering
  Data Encryption
  Database Management
  Computation by Abstract Devices
  Algorithm Analysis and Problem Complexity
  
• 出版者：Springer Berlin / Heidelberg
• ISSN：1611-3349

文摘

Object detection systems which operate on large data streams require an efficient scaling with available computation power. We analyze how the use of tile-images can increase the efficiency (i.e. execution speed) of distributed HOG-based object detectors. Furthermore we discuss the challenges of using our developed algorithms in practical large scale scenarios. We show with a structured evaluation that our approach can provide a speed-up of 30-180 % for existing architectures. Due to the its generic formulation it can be applied to a wide range of HOG-based (or similar) algorithms. In this context we also study the effects of applying our method to an existing detector and discuss a scalable strategy for distributing the computation among nodes in a cluster system.