Research & Development

Aged-care facilities needed a reliable way to detect in-bed, out-of-bed, and repositioning events to prevent falls and pressure injuries, but existing systems were inaccurate, untested in Australian conditions, and unsuitable for real-world care environments.

We developed a smart-mattress analytics system that uses a 240-point pressure sensor pad along with a novel behaviour-analytics algorithm. The system was refined through controlled data collection, infrastructure assessment, and live trials in aged-care facilities.

Delivered accurate event detection, stable wireless performance, and validated results across multiple care sites, demonstrating a clinically meaningful and operationally reliable fall and pressure-injury prevention tool.

Retail centres required demographic insights, but overhead CCTV made facial-based analytics impossible due to privacy concerns, angle, lighting variability and crowding. Existing AI systems could not operate reliably under these constraints.

We built a camera-agnostic AI model that determines gender using non-facial appearance cues, supported by custom preprocessing, lighting normalisation, occlusion handling, and frame-averaging techniques. The system runs on standard CCTV without specialised hardware.

Achieved high accuracy across diverse lighting conditions and camera types, maintained low false positives in crowded scenes, and was successfully validated in live shopping-centre environments.

Healthcare and aged-care environments needed a privacy preserving way to detect human actions such as presence, sitting, standing, and sleeping. Camera-based systems raised privacy concerns, wearables suffered from compliance issues, and existing WiFi-sensing research failed to perform reliably outside controlled lab conditions.

A WiFi-CSI-based activity-recognition system was developed using custom ESP32 hardware, proprietary signal-processing pipelines, engineered CSI features, and optimised machine-learning and deep-learning models. The system converts ambient WiFi signal variations into real-time activity insights without cameras or wearables.

Achieved ~90% classification accuracy, improved cross-position generalisation, reduced posture-confusion rates, and demonstrated sub-2-second detection latency. A large, anonymised CSI dataset was created to support future commercialisation and IP development.

Underground mines needed a fire-safe, intrinsically safe backup-power solution for critical electronics. Traditional battery-based UPS units posed ignition risks, degraded quickly, and were unsuitable for harsh mining environments.

A supercapacitor-based UPS system was engineered for intrinsic safety, long cycle life, and stable performance in high-temperature, high-vibration conditions. The design eliminated thermal and chemical risks while providing reliable short-term backup power.

Delivered a rugged, fire-safe power-backup solution that improves operational safety, reduces maintenance requirements, and ensures continuity of critical underground systems.