Proctor, Fern (2023) Real Time Drowning Detection System Using Machine Vision and Learning. [USQ Project]
![[img]](https://sear.unisq.edu.au/style/images/fileicons/text.png) |
Text (Project – redacted)
Proctor_F_Low_Redacted.pdf Download (5MB) |
Abstract
Globally, there are an estimated 236,000 drownings per year (World Health Organisation, 2014), and in Australia, drowning deaths form a grim tally over the summer months despite the introduction of robust legislation and education campaigns. Swimming pools are the most common location for drowning accidents (AIHW, 2023) and as such, any tools that may be able to reduce these occurrences could have substantial impact.
There have been significant advances in the fields of machine vision and learning in the past decade resulting in the development of new algorithms and hardware which can be applied to novel applications. Of particular interest is the ‘You Only Look Once’ (YOLO) algorithm which offers real time detection speeds with impressive accuracy. With a solution looking for a problem to solve, this project aimed to develop a real time drowning detection system using machine learning and vision. In particular, the system was designed to be used in a residential pool setting with little technical knowledge required for installation and setup.
A dataset was acquired and supplemented with additional images. Several versions of the Yolo algorithm were examined and trained using the custom dataset. Image augmentation and hyperparameter tuning were among some of the methods used to improve the model’s accuracy prior to it being deployed on an Oak-D Lite, an edge-AI camera where neural inference is done onboard rather than on a separate device. Finally, the system was deployed and tested in real time as well as monitored remotely through a mobile device. The testing demonstrated that real time drowning detection was feasible using YoloV8 and the Oak-D Lite.
Future works include improving and increasing the size of the dataset used for training as well as experimenting with various iterations of the Yolo algorithms deployed on the Oak-D Lite, in particular smaller versions to compare their accuracy with YoloV8-m used in the final system. In addition, further investigation of hyperparameter tuning and image augmentation could be beneficial as well as development of a mobile interface with the ability to notify users of a drowning event by means of a notification or via smart devices such as watches.
|
Statistics for this ePrint Item |
| Item Type: | USQ Project |
|---|---|
| Item Status: | Live Archive |
| Faculty/School / Institute/Centre: | Current – Faculty of Health, Engineering and Sciences - School of Engineering (1 Jan 2022 -) |
| Supervisors: | Low, Tobias |
| Qualification: | Bachelor of Engineering (Electrical and Electronic) |
| Date Deposited: | 01 Oct 2025 03:59 |
| Last Modified: | 01 Oct 2025 03:59 |
| Uncontrolled Keywords: | drownings; machine vision and learning |
| URI: | https://sear.unisq.edu.au/id/eprint/52993 |
Actions (login required)
 |
Archive Repository Staff Only |
