Indoor Human Behavior Recognition Method Based on Wavelet Scattering Network and Conditional Random Field Model

Xiaodong Qu*, Weicheng Gao, Haoyu Meng, Yi Zhao, Xiaopeng Yang

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

5 Citations (Scopus)

Abstract

Ultrawideband (UWB) through-the-wall radar (TWR) can be used for indoor human behavior recognition via micro-Doppler information. However, faced with weak micro-Doppler features and low accuracy, conventional recognition method does not perform well in shielded environment. To address these problems, this article proposes an indoor human behavior recognition method based on the wavelet scattering network and conditional random field model (TWR-WSN-CRF). In the proposed method, wavelet scattering network (WSN) and speckle reducing anisotropic diffusion (SRAD) with weighted guided image filter (WGIF) are used for feature enhancement and noise suppression, and the signal-to-noise ratio (SNR) is improved. Then, the human behavior recognition network based on the conditional random field (CRF) model is developed to extract global and local features from the wall, target, and noise subspaces obtained by singular value decomposition (SVD). Finally, the multilayer perceptron (MLP) model and weighted majority voting (WMVE) method are used for fusion decision. The effectiveness of the proposed method is verified by experiment. The results show that compared with other methods, the proposed human behavior recognition method achieves highest recognition accuracy with 96.25% on the validation dataset.

Original languageEnglish
Article number5104815
JournalIEEE Transactions on Geoscience and Remote Sensing
Volume61
DOIs
Publication statusPublished - 2023

Keywords

  • Conditional random field (CRF)
  • human behavior recognition
  • through-the-wall radar (TWR)
  • wavelet scattering network (WSN)

Fingerprint

Dive into the research topics of 'Indoor Human Behavior Recognition Method Based on Wavelet Scattering Network and Conditional Random Field Model'. Together they form a unique fingerprint.

Cite this