We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.LG

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Electrical Engineering and Systems Science > Signal Processing

Title: A MIMO Radar-Based Metric Learning Approach for Activity Recognition

Abstract: Human activity recognition is seen of great importance in the medical and surveillance fields. Radar has shown great feasibility for this field based on the captured micro-Doppler ({\mu}-D) signatures. In this paper, a MIMO radar is used to formulate a novel micro-motion spectrogram for the angular velocity ({\mu}-{\omega}) in non-tangential scenarios. Combining both the {\mu}-D and the {\mu}-{\omega} signatures have shown better performance. Classification accuracy of 88.9% was achieved based on a metric learning approach. The experimental setup was designed to capture micro-motion signatures on different aspect angles and line of sight (LOS). The utilized training dataset was of smaller size compared to the state-of-the-art techniques, where eight activities were captured. A few-shot learning approach is used to adapt the pre-trained model for fall detection. The final model has shown a classification accuracy of 86.42% for ten activities.
Comments: 6 pages, 7 figures, 2 tables
Subjects: Signal Processing (eess.SP); Machine Learning (cs.LG)
Cite as: arXiv:2111.01939 [eess.SP]
  (or arXiv:2111.01939v1 [eess.SP] for this version)

Submission history

From: Fady Aziz [view email]
[v1] Tue, 2 Nov 2021 23:11:53 GMT (4598kb,D)

Link back to: arXiv, form interface, contact.