Abstract:Activity recognition is one of the significant technologies accompanying the development of the Internet of Things (IoT). It can help in recording daily life activities or reporting emergencies, thus improving the user's quality of life and safety, and even easing the workload of caregivers. This study proposes a human activity recognition (HAR) system based on activity data obtained via the micro-Doppler effect, combining a two-stream one-dimensional convolutional neural network (1D-CNN) with a bidirectional gated recurrent unit (BiGRU). Initially, radar sensor data are used to generate information related to time and frequency responses using short-time Fourier transform (STFT). Subsequently, the magnitudes and phase values are calculated and fed into the 1D-CNN and Bi-GRU models to extract spatial and temporal features for subsequent model training and activity recognition. Additionally, we propose a simple cross-channel operation (CCO) to facilitate the exchange of magnitude and phase features between parallel convolutional layers. An open dataset collected through radar, named Rad-HAR, is employed for model training and performance evaluation. Experimental results demonstrate that the proposed 1D-CNN+CCO-BiGRU model demonstrated superior performance, achieving an impressive accuracy rate of 98.2%. This outperformance of existing systems with the radar sensor underscores the proposed model's potential applicability in real-world scenarios, marking a significant advancement in the field of HAR within the IoT framework.
engineering, electrical & electronic,chemistry, analytical,instruments & instrumentation
What problem does this paper attempt to address?
The problem that this paper attempts to solve is to improve the accuracy and applicability of human activity recognition (HAR) by using radar data based on the micro - Doppler effect and deep - learning techniques within the Internet of Things (IoT) framework. Specifically, the researchers proposed an HAR system that combines two - stream one - dimensional convolutional neural network (2S 1D - CNN) and bidirectional gated recurrent unit (BiGRU), aiming to recognize users' daily activities through data obtained from radar sensors. These activities include boxing, clapping, waving, walking, running, and standing still.
### Research Background and Motivation
With the intensification of population aging, the demand for elderly care is increasing day by day. Traditional care methods often rely on manpower, which are not only inefficient but also prone to negligence. Therefore, it has become particularly important to develop technologies that can monitor and recognize the daily activities of the elderly in real - time. HAR technology can collect user movement information through various sensors and recognize users' activities according to unique movement patterns, thereby providing real - time monitoring and helping to determine whether users' daily activities are normal. For example, for healthy users, daily exercise and sleeping times can be set, and reminders can be sent out through IoT devices; in the event of an accident, the system can notify family members and medical units to prevent treatment delays.
### Research Methods
1. **Dataset**: The research used the Rad - HAR public dataset, which was provided by the Temple University Advanced Signal Processing Laboratory. The Ancortek SDR - KIT 2500B radar was used to collect information on eight user actions. These actions include boxing, clapping, waving, walking in place, running towards the radar, running away from the radar, walking towards the radar, and walking away from the radar. Finally, the model classified running towards the radar and running away from the radar as "running" activities, and walking towards the radar and walking away from the radar as "walking" activities, for a total of six different activities.
2. **Data Pre - processing**:
- Use MATLAB to down - sample all data to 64 K samples per second.
- Remove data close to the DC component because these data are unrelated to actions and will interfere with model performance.
- Apply the short - time Fourier transform (STFT), select the Hamming window and set an overlap rate of 90% to obtain the micro - Doppler spectrogram.
- Calculate the magnitude and phase of STFT as follows:
\[
A_{tf}=\sqrt{S_r(\omega)^2 + S_i(\omega)^2}
\]
\[
\phi_{tf}=\tan^{-1}\left(\frac{S_i(\omega)}{S_r(\omega)}\right)
\]
- Store the magnitude and phase in complex form, where the real part is the magnitude and the imaginary part is the phase:
\[
x_{tf}=A_{tf}+j\phi_{tf}
\]
- Use the sliding - window technique to segment the data, with each sample containing 75 time frames (corresponding to 1 second) and an overlap of 25 time frames between adjacent samples.
3. **Model Architecture**:
- **Spatial Feature Extraction Unit**: Consists of four layers of 2S 1D - CNN + CCO, used to extract spatial features.
- **Temporal Feature Extraction Unit**: Consists of two Bi - GRU, with the magnitude feature and the phase feature input respectively.
- **Fusion Layer**: After splicing the spatial features and the temporal features, perform action recognition through three fully - connected layers.
### Experimental Results
The experimental results show that the proposed 2S 1D - CNN + CCO - BiGRU model achieved an accuracy of 98.2% on the Rad - HAR dataset, significantly outperforming existing systems. This indicates that the model has high application potential in practical scenarios, especially in HAR tasks within the IoT framework.
### Conclusion
This research proposed an efficient HAR system by combining radar sensor data and deep - learning techniques, which can achieve high - precision recognition in a variety of daily activities. This achievement not only helps to improve the quality of life and safety of the elderly, but also paves the way for future smart living and health management.