COMP SCI 4092/7092: Mobile and Wiresless Systems

Watchdog Project. This project-based-learning (PBL) course requires a group of students (3-4 students in one group) to develop a mobile platform for an accurate and reliable method for addressing wandering-off using an embedded camara system (e.g., a Raspberry Pi 4 with a webcam).

Background

Hospitals and residential homes have a significant need for monitoring and recognising wandering-off (e.g. elopement) older people with cognitive impairments because of the severe consequences arising from wandering-off such as disappearances and serious injuries, for example, from collisions with vehicles in parking lots.

Problem:

Due to increasing ageing populations across the globe, we can expect wandering-off to become a significant problem of scale affecting all of us. Existing technologies used to address wandering-off are inadequate for providing close supervision as they use proximity-based sensing that often leads to false alarms.

Description:

You will be working to develop localisation and tracking algorithms for mobile and wireless systems, implementing these using an embedded camera system (e.g. a Raspberry Pi system with a camera) to develop a system capable of being customised to monitor multiple individuals across specific thresholds.

In particular, we want to build a system that detects and tracks a person entering and leaving a room and their path using a single camera. Our approach uses YOLOv3 to identify people and then using a Kalman filter (KF) based technique to track people continuously. Using real-time spatial and temporal data obtained from the KF based tracking algorithm, we develop two algorithms: i) person traversing direction (TD) algorithm to identify the person’s moving direction (e.g. moving out of a room), and ii) person traversing path detection algorithm (TPD) to estimate the traversal path of that person.

The watchdog platform has two parts: i) server; and ii) mobile application (web app). If a forbidden path or a movement, for example, predefined by a caregiver for an individual patient, is identified, then an alert message is sent to the caregiver’s smartphone. The smartphone app should then provide additional details such as the person’s name, location, time, traversal path and traversal direction on a floor plan. Monitoring rules are not the same for all patients. Some patients may be allowed to go beyond a given threshold, while other specific patients may not be. Therefore the system should be able to identify movements that are not allowed for individual patients and correctly send an alert message through a caregiver. The same mobile app should have an interface for senior caregivers to enter new patients, removing new patients as well as setting rules for those patients to control alerting.

Outcomes:

Minimum performance:

• A working prototype system for demonstration at the end of the project (at least with a Phone App and a Pi with a camera).
• The system can at least track a person and trigger an alarm to the caregiver.

Above Minimum Performance:

• A cloud-based system (e.g. A raspberry Pi + AWS + A smartphone).
• A system that can identify the patient’s identity correctly.
• A high quality experiment and results.