Related Experiment Video
Updated: Jul 13, 2025
11:54
Microfluidic Platform with Multiplexed Electronic Detection for Spatial Tracking of Particles
Published on: March 13, 2017
9.3K
A Multitask Network for People Counting, Motion Recognition, and Localization Using Through-Wall Radar
Junyu Lin1, Jun Hu1, Zhiyuan Xie1
1School of Electronics and Communication Engineering, Sun Yat-sen University, Shenzhen 518107, China.
Sensors (Basel, Switzerland)
|October 14, 2023
Summary
This study introduces a novel multitask network for through-wall radar (TWR) systems. The network accurately performs people counting, action recognition, and localization simultaneously, enhancing safety and rescue operations.
Area of Science:
- Radar Systems Engineering
- Artificial Intelligence in Security
- Signal Processing
Background:
- Through-wall radar (TWR) utilizes low-frequency electromagnetic waves for penetrating detection, crucial for public safety, counterterrorism, and disaster rescue.
- Current TWR research often focuses on single tasks like detection or counting, limiting comprehensive situational awareness.
- Practical TWR applications demand simultaneous people counting, action recognition, and precise localization.
Purpose of the Study:
- To develop and validate a multitask network capable of concurrently executing people counting, action recognition, and localization using TWR data.
- To process range-time-Doppler (RTD) spectra from 1D radar signals for comprehensive human activity analysis.
- To address the limitations of single-task TWR systems by offering integrated people monitoring capabilities.
Main Methods:
- A multitask deep learning network employing convolutional layers and attention modules was designed to process RTD spectra.
- Confidence matrices were generated, encoding people count, motion category, and location information as labels.
- A summed individual task loss function was formulated, converting the localization problem into a multilabel classification task.
Main Results:
- The multitask network achieved 96.94% accuracy in people counting and 96.03% in motion recognition on a test set of 10,032 samples.
- The system demonstrated an average distance error of 0.12 m for localization through a 24 cm thick brick wall.
- The proposed method effectively extracts deep features for simultaneous multi-task TWR analysis.
Conclusions:
- The developed multitask network offers a robust and accurate solution for integrated people counting, action recognition, and localization in through-wall scenarios.
- This approach significantly advances the capabilities of TWR systems for enhanced public safety and emergency response.
- The findings highlight the potential of deep learning and attention mechanisms in complex radar signal processing for real-world applications.
