Hopkins, Andrew (2020) Pre-Emptive Detection of Mobile Phone Use In-Vehicle. [USQ Project]
|
Text (Project)
Hopkins_A_Low_Redacted.pdf Download (4MB) | Preview |
Abstract
Distraction while driving currently makes up 14% of crashes and 10% of fatalities on roads in NSW, Australia on average every year (Transport for NSW 2020a). Through the increase in reliance on mobile devices, these numbers are sure to increase. There are currently methods involved in the reduction of this issue, however there is currently nothing in place for detection while driving.
This research works to prove the hypothesis that using a fixed camera installed in a car, and with some image pre-processing and machine learning algorithms applied, a machine can detect the act of reaching for a phone. To prove this hypothesis, the research required a database to be created, and algorithms used in order to determine the accuracy of detection.
With the analysis of previous literature, this research was able to identify some key methods required for classification of pre-emptive mobile phone usage. These techniques were tested against a database created with varying vehicle types and driver behaviour. The results of the trained models were then tested against a previously ‘unseen’ data set, to verify the accuracy of the machine.
The results of this project showed promise, with three different machine learning techniques applied against unseen data. Each reviewed 240 samples of reaching events as well as a varied length of driving videos. The bagged tree method detecting 27.92% with a false positive rate of 3.45 per minute, the cosine KNN method detecting 29.17% with a false positive rate of 4.74 per minute and the cubic SVM method detecting 29.17% with a false positive rate of 6.13 per minute.
This research was limited by hardware capabilities, as well as data limitations. With an increased database across multiple vehicle types and drivers, the results would show more significance. Further research in this area would also be required to limit the time of which the pre-processing algorithms required, as 1 second of data is still currently requiring 2.8 seconds in processing time.
Suggestions were made that additional research with a more varied database should be conducted to ensure validity across different vehicle types and driving styles. With more significant training conducted, a machine could then be developed in order to make these calculations real time and test them against a driver in a vehicle. This technique would show significant importance in detecting the shortfalls of the machine training and identifying different areas for potential improvement. Additional trials on pre-processing techniques were identified and some ideas around the training to identify more than just three categories to help in determining areas of which data is lacking.
|
Statistics for this ePrint Item |
| Item Type: | USQ Project |
|---|---|
| Item Status: | Live Archive |
| Faculty/School / Institute/Centre: | Historic - Faculty of Health, Engineering and Sciences - School of Mechanical and Electrical Engineering (1 Jul 2013 - 31 Dec 2021) |
| Supervisors: | Low, Tobias |
| Qualification: | Bachelor of Engineering (Honours) (Electrical and Electronic) |
| Date Deposited: | 23 Aug 2021 04:56 |
| Last Modified: | 26 Jun 2023 04:05 |
| URI: | https://sear.unisq.edu.au/id/eprint/43061 |
Actions (login required)
 |
Archive Repository Staff Only |
