AI-powered radar for emergency response

Justin Kim, Luc McDonald, Lakshya Saroha, Louie Tang and Yuheng Wu

Community Partner: Delta Controls
Degree:
- Bachelor of Applied Science
Program:
- Electrical and Computer Engineering
Campus: Vancouver

Our project

Medical staff caring for patients in seclusion rooms face a difficult monitoring problem. These patients are often sedated or on medications that increase the risk of cardiac or respiratory emergencies, as well as falls. Traditional vital‑sign monitoring tools require physical contact and introduce safety concerns because the patient could use the wires to harm themselves or others. In current practice, staff conduct manual checks every 15 minutes, which is labour‑intensive and still leaves gaps when a medical emergency could go unnoticed.

Our project addresses this challenge by developing a fully contactless system that can detect falls and monitor heart rate and respiratory rate in real time. The system is designed to immediately alert medical staff if it detects abnormal vital signs or a fall, reducing both the workload on staff and the risk of unobserved emergencies.

Delta Controls

Our design solution

We developed an AI-powered mmWave radar that detects a patient’s heart and respiratory rate entirely through contactless sensing. The radar operates in real time, detecting micro‑displacements of the chest surface, typically between 0.1 and 0.5 mm, to estimate heart rate. Because these signals are extremely sensitive to movement, we integrated a thermal sensor that determines whether the patient is localized within the bed area. When the thermal sensor confirms the patient’s presence, the system automatically begins monitoring for vital signs.

We developed a signal processing algorithm, that extracts the patient’s heart rate and respiratory rate from their chest movement with less than 4 BPM and 2 RPM average accuracy. Abnormal heart rate patterns are detected and flagged using an autoencoder machine learning model, and dangerously high or low heart rates and respiratory rates are flagged by a configurable threshold check.

To detect falls, we developed a separate signal processing algorithm that extracts macro-movements from the incoming radar data. We trained a convolutional neural network on the processed macro-movements, with testing showing it can detect falls to 97.93% accuracy.

All sensor data are managed by a central server and integrated into a web application that displays information in real time so medical staff can respond as needed. The entire system is housed in a custom ceiling-mounted assembly.

AI-powered Radar Project Capstone Poster.

See our brochure for more details!

The technical challenges we faced

One of the biggest technical challenges was developing the vital‑sign component of the system. The underlying signal‑processing research for mmWave radar is still emerging, and very little of the published work demonstrates the level of accuracy we needed for our project.

Needless to say, it is challenging to extract a heartbeat based on skin displacement of 0.1 to 0.5 millimetres, particularly with the larger displacement of breathing layered on top of that, and separating those two signals!

We also needed our solution to work in real time, which is not easily supported by the mmWave radar we are using. We built a workaround that allowed us to stream and update vital‑sign estimates every second.

How we validated our solution

We validated our system in two ways. When we were developing the individual components, such as the fall‑detection module and the vital‑sign extraction module, we tested each piece using publicly available datasets to confirm that our algorithms behaved as expected.

After we integrated the two modules into a single system, we shifted to real-world testing.

We were our own test subjects: we simulated falls, lay in front of the radar to evaluate vital signs and observed how the system responded in real time.

What we’re most proud of

We came into engineering to find a way to use our skills to make a difference in the world. This project does that by addressing an actual problem in the realm of health care. When we’ve talked with health-care practitioners at various stages of this project they have told us how useful this kind of product could be.

This complex project let us put into practice something we’ve learned since first year: that one of the primary duties of an engineer is to consult with stakeholders and translate their needs into practical, reliable solutions.

We were able to integrate seemingly independent sub-system components into a single product that can be deployed in a hospital and change people’s lives – making it easier for health care providers to provide care and for patients to stay safe. There could be broader applications as well, such as monitoring the well-being of seniors.

Finally, a key aspect of engineering is teamwork. Throughout this project, we functioned extremely well as a collaborative and respectful team.

We each applied our own expertise and helped each other out in areas where we were not as strong so that we could all develop new skills, learn from each other and deliver something meaningful together.