A new CQUniversity-led study has investigated a machine learning approach to detect sitting and sleep history in drivers' with the aim to reduce the number of fatigued and sleepy drivers on the road.
CQU Doctor of Philosophy (PhD) student Georgia Tuckwell conducted the study alongside a predominantly female research team' Dr Charlotte Gupta' Dr Grace Vincent and Professor Sally Ferguson from CQUniversity's Appleton institute' highlighting the significant impact of women who are largely under-represented in this field of research.
The cross-institutional team also consisted of researchers James Keal from the University of Adelaide and Jarrad Kowlessar from Flinders University.
"This research is targeted at helping to identify at-risk drivers and reduce the impact of fatigue on the road'" Ms Tuckwell said.
"Fatigued driving as a result of inadequate sleep' poses a significant risk to drivers and is a major factor in 20 per cent of all road accidents worldwide.
"Previous research clearly demonstrates major differences in driving performance between drivers who obtain an adequate or inadequate amount of sleep per night.
"However' what was unknown was whether we could classify driver sleep history solely from physical movements during driving using sensors placed on the body."
The Adelaide-based researcher explained that the research successfully utilised a multidisciplinary machine learning approach which differed from previous driver fatigue classification studies which used computer vision information derived from images.
"Since our study utilised body-worn sensors capturing movement as it pertains to the control of the vehicle' we started to explore how we could apply a deep learning approach using only time-series data' instead of images' to classify changes in movement as a result of inadequate sleep'" Ms Tuckwell explained.
"The body-worn accelerometer device was attached to the right thigh to provide a snapshot of the movement involved in the control of the vehicle' such as braking and accelerating.
"Two different deep learning networks' called convolutional neural networks' were utilised in this study' and we found that our custom implementation of a program called ResNet-18' was able to achieve the highest accuracy rate to classify sleep history.
"This network was able to accurately classify if a driver obtained an adequate (nine hours) or inadequate (five hours) amount of sleep' 88 per cent of the time. This accuracy rate was achieved from less than four minutes of data.
"This application of deep learning using body-worn sensors may help to provide an alternate means to current computer vision technologies for future driver fatigue and impairment detection."
This study will form one part of Ms Tuckwell's thesis' which she hopes will demonstrate what can be achieved when combining artificial intelligence and body-worn sensors for detecting at-risk driving behaviours.
"My hope is that in the future we will be able to identify a fatigued or impaired driver within the smallest time frame possible!
"I plan to continue work in driver safety research and work within multidisciplinary teams to further explore how deep learning can make us all safer drivers on the road'" she said.
"This data was collected within a larger study with several experts from CQU's Appleton Institute. This larger study looks at the cognitive and health outcomes of breaking up sitting during the day for sedentary workers."
To help share her research' Ms Tuckwell recently entered CQU's annual Visualise Your Thesis (VYT) competition' winning the 60-second VYT challenge with her animated video presentation.
"It was a great experience and allowed me to share my research with a wider audience' particularly how machine learning can be utilised in research.
"I am passionate about seeing more women become more strongly represented in previously under-represented areas of research'" she said.
Watch Georgia Tuckwell's VYT entry here - most-viewed entry will win the Trending on VYT challenge!
"I only began to teach myself programming six months into my PhD and wished that this was an area I had been exposed to' and felt more welcome in earlier in my career.
"Throughout my career' I would love to support other women and non-binary researchers in traditionally male-dominated research areas and spaces."