Enhancing Vital Sign Estimation Performance of FMCW MIMO Radar by Prior Human Shape Recognition

Hadi Alidoustaghdam,Min Chen,Ben Willetts,Kai Mao,André Kokkeler,Yang Miao

2023-03-16

Abstract:Radio technology enabled contact-free human posture and vital sign estimation is promising for health monitoring. Radio systems at millimeter-wave (mmWave) frequencies advantageously bring large bandwidth, multi-antenna array and beam steering capability. \textit{However}, the human point cloud obtained by mmWave radar and utilized for posture estimation is likely to be sparse and incomplete. Additionally, human's random body movements deteriorate the estimation of breathing and heart rates, therefore the information of the chest location and a narrow radar beam toward the chest are demanded for more accurate vital sign estimation. In this paper, we propose a pipeline aiming to enhance the vital sign estimation performance of mmWave FMCW MIMO radar. The first step is to recognize human body part and posture, where we exploit a trained Convolutional Neural Networks (CNN) to efficiently process the imperfect human form point cloud. The CNN framework outputs the key point of different body parts, and was trained by using RGB image reference and Augmentative Ellipse Fitting Algorithm (AEFA). The next step is to utilize the chest information of the prior estimated human posture for vital sign estimation. While CNN is initially trained based on the frame-by-frame point clouds of human for posture estimation, the vital signs are extracted through beamforming toward the human chest. The numerical results show that this spatial filtering improves the estimation of the vital signs in regard to lowering the level of side harmonics and detecting the harmonics of vital signs efficiently, i.e., peak-to-average power ratio in the harmonics of vital signal is improved up to 0.02 and 0.07dB for the studied cases.

Signal Processing

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: how to enhance the performance of FMCW MIMO radar in estimating vital signs (such as respiration rate and heart rate) through pre - recognized human body shapes and postures. Specifically, the paper aims to solve the following two key problems:

1. **Sparse and Incomplete Human Point - Cloud Data**:
- The human point - cloud data generated by millimeter - wave (mmWave) radar are usually sparse and incomplete, which makes it difficult to perform pose estimation based on these data.
- Random body movements (RBMs) will further deteriorate the estimation accuracy of vital signs. Therefore, it is necessary to accurately locate the chest and use narrow - beam radar to improve the estimation accuracy.

2. **Improving the Accuracy of Vital Sign Estimation**:
- By using convolutional neural networks (CNN) to process imperfect point - cloud data to recognize human postures and extract key information of the chest, more accurate vital sign estimation can be achieved.
- By using beam - forming technology to focus the radar beam on the chest area, the interference caused by non - informative body movements can be reduced, thereby improving the estimation performance of respiration rate (BR) and heart rate (HR).

### Overview of the Solution

To solve the above problems, the paper proposes a new pipeline method, which mainly includes the following steps:

1. **Human Pose and Shape Recognition**:
- Use a trained convolutional neural network (CNN) to efficiently process imperfect human point - cloud data and output key points of different body parts.
- The CNN framework is trained through RGB image reference and augmented ellipse fitting algorithm (AEFA) to provide more accurate pose estimation.

2. **Vital Sign Estimation**:
- Use the chest information in the previously estimated human pose for vital sign estimation.
- By using beam - forming technology to focus the radar beam on the chest, the signal - to - noise ratio and interference suppression ability of vital sign signals can be improved.

### Main Contributions

1. **Improving Contactless Human Pose Recognition**:
- Propose to use CNN to estimate human poses and model human shapes, where the shape contains size information of various body parts.
- Apply the AEFA algorithm to automatically fit ellipses on RGB images as the ground - truth data for training CNN.

2. **Demonstrating the Benefits of Pose Estimation for Vital Sign Extraction**:
- By using beam - forming technology to focus the radar beam on the chest position estimated by CNN, the performance of vital sign estimation can be significantly improved, especially the ratio of harmonic peak to average power (PAPR), with improvements of 0.02 dB and 0.07 dB respectively.

### Formula Representation

- The formula for mean - square error (MSE) is used as the loss function in the CNN training process:
\[
\epsilon_t(i)=\frac{1}{2}\|\tilde{C}_t^i - C_t^i\|^2_2
\]
where \(\tilde{C}_t^i\) represents the coordinate of the \(i\) - th estimated key point in the \(t\) - th frame, and \(C_t^i\) represents the corresponding ground - truth data.

- The definition of the loss function \(\epsilon_t\):
\[
\epsilon_t=\frac{1}{N_p}\sum_{i = 1}^{N_p}\epsilon_t(i)
\]

Through this method, the paper successfully demonstrates how to combine advanced machine - learning techniques and millimeter - wave radar techniques to improve the accuracy of vital sign monitoring.