TeleHealthMonitor: Edge AI for remote patient monitoring (CamvsCovid)
During the early 2020 COVID-19 surge, Indonesian hospitals needed to monitor high-risk patients at home. Standard systems required broadband and medical devices. These included pulse oximeters for blood oxygen and ECG patches for heart activity. However, rural broadband access was below 30%, and medical hardware supply chains were disrupted.
Analysing breathing on servers required video that would overwhelm rural 2G and 3G links. Therefore, analysis had to run locally. Low-end Android phones from 2017–2019 needed model compression and careful runtime tuning. Moreover, privacy rules barred raw patient video from leaving the device. This made local inference essential.
I designed the system to analyse breathing entirely on the device. A lightweight pose model located the chest. Then optical flow tracked chest movement between video frames to estimate breathing rate. The dashboard received only the derived vital-sign estimate as a single floating-point number. Therefore, video frames stayed on the device. Moreover, a structured mobile form collected symptoms to add clinical information.
I used quantisation, which stores model values at lower precision, and optimised inference for ARM CPUs. Then I tested FP32, FP16, and INT8 number formats on physical devices. These tests compared inference speed and accuracy as numerical precision changed.
I designed the system for GDPR compliance, including protection of personal data. To limit exposure, zero raw biometric data left the device. Transmitted data contained only aggregated vital-sign estimates and structured symptom reports.
The project placed in the global top 3 at Cambridge’s COVID-19 innovation challenge. It showed that privacy-compliant vital-sign monitoring could run on ordinary 2019 phones over 2G. Moreover, the design applies to other health deployments with strict bandwidth and hardware limits.