BiRAMNet: a selective lightweight model with bi-level routing attention for indoor human activity recognition

Yongwei Zhang; Yuan He; Xinqi Zhang

doi:10.1117/12.3022710

28 March 2024 BiRAMNet: a selective lightweight model with bi-level routing attention for indoor human activity recognition

Yongwei Zhang, Yuan He, Xinqi Zhang

Author Affiliations +

Proceedings Volume 13091, Fifteenth International Conference on Signal Processing Systems (ICSPS 2023); 130911V (2024) https://doi.org/10.1117/12.3022710
Event: Fifteenth International Conference on Signal Processing Systems (ICSPS 2023), 2023, Xi’an, China

Abstract

Indoor human activity recognition, when deployed on edge devices, holds promising application prospects in security surveillance. However, conventional CNN-based lightweight recognition methods perform poorly in indoor environments due to wall interference and signal attenuation. To address this issue, this paper proposes an effective lightweight model that balances parameters and performance to recognize seven types of human activities occurring behind walls. Specifically, the BiRAMNet consists of four pyramid-like stages and utilizes the Bi-level Routing and Residual (BRR) block, a fundamental block combining the strengths of CNNs and Transformers, as its building block. We introduce Bi-level Routing Attention to focus on the motion characteristics of the Doppler spectrogram, sidelining background information. The proposed model is benchmarked against baseline networks, including ResNet50, ResNet101, MobileNetv2, MobileNetv3 and EMO. The results highlight that our model achieves superior recognition accuracy and a lighter size. Future work includes extending BiRAMNet to more diverse human activity recognition scenarios and improving its performance for real-world applications.

(2024) Published by SPIE. Downloading of the abstract is permitted for personal use only.

Citation Download Citation

Yongwei Zhang, Yuan He, and Xinqi Zhang "BiRAMNet: a selective lightweight model with bi-level routing attention for indoor human activity recognition", Proc. SPIE 13091, Fifteenth International Conference on Signal Processing Systems (ICSPS 2023), 130911V (28 March 2024); https://doi.org/10.1117/12.3022710

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available