4th WCSET-2015 at Japan

 

Applied Sciences and Engineering:

Title:           A Single Depth Sensor based Human activity recognition via Deep Belief Network
Authors:      S. B. Nam, S. U. Park, J. H. Park, T. S. Kim
Abstract:     Human activity recognition (HAR) from videos is a challenging work towards various important application domains, such as human-machine interaction, interactive entertainment, smart surveillance, etc. In this paper, we present a novel system to recognize human activities via Deep Belief Network (DBN). We have employed Restricted Boltzmann Machine (RBM) over the commonly used Hidden Markov Models (HMMs). We train RBM with the spatio-temporal features of human activities (i.e, pre-training) via a contrastive divergence algorithm. This pre-training step makes the network avoid an over-fitting problem of the classical neural networks. Once the networks are pre-trained, we utilize a back propagation algorithm with known labels of the activity training sets to fine tune the recognizer which is later used for HAR. In order to verify our system, we have compared the results of proposed methodology with the HMM based HAR using a database of Microsoft Research Cambridge-12 (MSRC-12). Our experimental results show that the proposed approach is able to recognize various human activities and it outperforms HMMs. We have achieved the average recognition accuracy of 97.54% for 12 activities. The results are 5.05% more accurate than that of the HMM based HAR and 60 times faster in training time. Then we have implemented and tested our proposed system for real-time HAR. In our real-time system, we obtain sequential frames of human depth silhouettes captured by a single depth camera. Then, we utilize our trained Random Forests (RFs) to recognize the 31 human body parts from the depth silhouettes. From the body parts map, we extract a set of activity features which is a vector of joint angles from the recognized body parts from 50 sequential depth frames. Finally, we apply the activity features to the proposed DBN. We have verified the feasibility of HAR in real-time.

Keywords:   Human Activity Recognition, Depth Sensor, Deep Learning, Deep Belief Network, Restricted Boltzmann Machine
Pages:         015-019

Download Full Paper

  1. Pengeluaran SDY
  2. https://ifa2020.org/
  3. https://www.hopepartnershipforeducation.org/
  4. https://www.bencomo.org/
  5. https://rvic.org/
  6. https://ladiosabuenosaires.com/
  7. DATA SGP
  8. https://www.bathconsultancygroup.com/
  9. https://climateinternational.org/
  10. https://www.tadmc.org/
  11. https://www.innvision.org/
  12. KELUARAN HK
  13. https://www.eccsit.org/
  14. KELUARAN HK
  15. https://www.la-boissaude.com/
  16. https://www.runnertriathletenews.com/
  17. https://www.lesfilmsbiographiques.com/
  18. DATA SDY
  19. https://www.shepherdoftheridge.org/
  20. https://www.musindioufu.org/
  21. Togel Hongkong
  22. https://everstribute.org/
  23. https://cosac-ndt.com/
  24. https://www.ederna.com/
  25. https://civicinnovationni.org/
  26. https://www.prattkan.com/
  27. KELUARAN SGP
  28. https://www.paganconference.com/
  29. https://francejpo.com/
  30. https://elk-mountain.com/
  31. https://www.globalteachinginsights.org/
  32. https://hwsevents.com/
  33. https://www.wildfireprograms.com/
  34. https://www.ibd-association.com/
  35. https://www.basharesearch.com/
  36. https://casori.org/
  37. https://www.tvemf.org/
  38. https://greatrough.com/
  39. https://www.nuitsdesarenes.com/
  40. https://www.tgssa.org/
  41. https://escapadesetflaneries.com/
  42. https://www.austenquotes.com/
  43. https://www.pahcs.com/
  44. https://www.galileoenrichment.com/