Dataset of human motion status using IR-UWB through-wall radar

Zhengliang Zhu, Degui Yang, Zhang Junchao, Feng TongJournal of Systems Engineering and Electronics 2021 · DOI

Abstract

Ultra-wideband (UWB) through-wall radar has a wide range of applications in non-contact human information detection and monitoring. With the integration of machine learning technology, its potential prospects include the physiological monitoring of patients in the hospital environment and the daily monitoring at home. Although many target detection methods of UWB through-wall radar based on machine learning have been proposed, there is a lack of an open-source dataset to evaluate the performance of the algorithm. This published dataset was measured by an impulse radio UWB (IR-UWB) through-wall radar system. Three test subjects were measured in different environments and several defined motion statuses. Using the presented dataset, we propose a human-motion-status recognition method using a convolutional neural network (CNN). On the well-trained network, the recognition accuracy of testing data for three kinds of motion statuses is higher than 99.7%.

Introduction

Through-wall radar imaging (TWRI) is a critical technology for post-disaster rescue, anti-terrorism, and non-contact health monitoring. Unlike optical cameras, radar respects privacy and can operate in total darkness or through smoke and debris. However, the field currently faces a significant hurdle: a lack of open-source, standardized datasets. Most researchers build proprietary hardware and collect private data, making it nearly impossible to compare the performance of different machine learning algorithms on a level playing field. This paper addresses this gap by providing a public dataset and a robust baseline model for human motion status recognition.

Experimental Methodology

The researchers employed an independently designed IR-UWB radar system with a center frequency of 500 MHz and a pulse repetition frequency of 400 KHz. This low frequency was specifically chosen for its ability to penetrate non-metallic obstacles like brick walls. The study defined three primary motion statuses for classification: human walking, human standing still, and an empty scene.

Schematic diagram of the simulation scene showing (a) human walking trajectories and (b) human standing still within the radar line of sight.

Instead of using complex micro-Doppler spectrograms, the team focused on time-domain range profiles. These are 2D images (768x32 pixels) where one axis represents distance (range) and the other represents time (slow time). Each 4-second sample captures the unique "signature" of a specific motion.

Dataset and CNN Architecture

The IR-UWB Human Motion Status Dataset includes 4,230 time-domain range profile samples collected from three healthy adult volunteers. Measurements were taken in both "Free Space" (no obstacles) and "Through-Wall" (0.2m thick brick-concrete wall) conditions.

To classify these statuses, a CNN inspired by the VGGNet architecture was designed. It consists of four convolutional layers with increasing filter counts (16 to 128), max-pooling layers, and two fully connected layers. The final output uses a Softmax classifier to categorize the motion into one of the three defined classes.

Key Results and Insights

The CNN model demonstrated exceptional performance, achieving an overall accuracy of 99.7% on the test set. Notably, the model achieved 100% accuracy for identifying "Human Walking" and "Empty Scene."

Key findings from the study include:

  • Feature Invariance: While different media (like brick walls) distort signal amplitude and propagation speed, the fundamental features of the motion status remain invariant. This suggests that models trained on free-space data can generalize effectively to through-wall scenarios.
  • Time-Window Trade-offs: Longer observation windows provide more detail for classification but reduce real-time efficiency. If a window is too long, overlapping motion states can confuse the classifier.
  • Clustering: Visualization using t-SNE (t-distributed Stochastic Neighbor Embedding) showed that the three motion statuses form distinct, non-overlapping clusters, validating the CNN's feature extraction capabilities.

Conclusion and Future Work

While the current results are highly accurate, the authors acknowledge that real-world rescue scenarios involve greater complexity, such as multiple people and varied wall materials. Future work will explore instantaneous feature extraction to enable true real-time monitoring and multi-dimensional feature extraction to recognize more fine-grained activities beyond simple walking and standing. The authors call on the UWB radar community to collaborate on building larger, more complex open-source datasets to further advance the field.

Summary updated

Abstract

Ultra-wideband (UWB) through-wall radar has a wide range of applications in non-contact human information detection and monitoring. With the integration of machine learning technology, its potential prospects include the physiological monitoring of patients in the hospital environment and the daily monitoring at home. Although many target detection methods of UWB through-wall radar based on machine learning have been proposed, there is a lack of an opensource dataset to evaluate the performance of the algorithm. This published dataset is measured by impulse radio UWB (IR-UWB) through-wall radar system. Three test subjects are measured in different environments and several defined motion status. Using the presented dataset, we propose a human-motion-status recognition method using a convolutional neural network (CNN), and the detailed dataset partition method and the recognition process flow are given. On the well-trained network, the recognition accuracy of testing data for three kinds of motion status is higher than 99.7%. The dataset presented in this paper considers a simple environment. Therefore, we call on all organizations in the UWB radar field to cooperate to build opensource datasets to further promote the development of UWB through-wall radar.
BibTeX
@article{zhu2021dataset,
  author = {Zhengliang Zhu and Degui Yang and Zhang Junchao and Feng Tong},
  title = {{Dataset of human motion status using IR-UWB through-wall radar}},
  journal = {Journal of Systems Engineering and Electronics},
  year = {2021},
  doi = {10.23919/jsee.2021.000093},
}
Figure