Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Driver phone use detection using visual, audio and inertial sensor processing
Dissertation (MEng (Computer Engineering))--University of Pretoria, 2019
Saved in:
| Other Authors: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
University of Pretoria
2020
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| _version_ | 1867613524137607168 |
|---|---|
| access_status_str | Open Access |
| author2 | Myburgh, Hermanus Carel |
| author_browse | Myburgh, Hermanus Carel |
| author_facet | Myburgh, Hermanus Carel |
| collection | Thesis |
| dc_rights_str_mv | © 2019 University of Pretoria. All rights reserved. The copyright in this work vests in the University of Pretoria. No part of this work may be reproduced or transmitted in any form or by any means, without the prior written permission of the University of Pretoria. |
| description | Dissertation (MEng (Computer Engineering))--University of Pretoria, 2019 |
| format | Thesis |
| id | oai:repository.up.ac.za:2263/73454 |
| institution | University of Pretoria (South Africa) |
| language | English |
| last_indexed | 2026-06-10T12:37:30.755Z |
| license_str | Other — see source repository |
| provenance_str_mv | Harvested via OAI-PMH from UPSpace — University of Pretoria Institutional Repository |
| publishDate | 2020 |
| publishDateRange | 2020 |
| publishDateSort | 2020 |
| publisher | University of Pretoria |
| publisherStr | University of Pretoria |
| record_format | dspace |
| source_str | UPSpace — University of Pretoria Institutional Repository |
| spelling | oai:repository.up.ac.za:2263/73454 Driver phone use detection using visual, audio and inertial sensor processing Myburgh, Hermanus Carel u12060055@tuks.co.za Viviers, Barend Jacobus Phone use detection Phone localisation Driver distraction Behaviour classification Talking on the phone and texting detection UCTD Dissertation (MEng (Computer Engineering))--University of Pretoria, 2019 Driver distraction is a major cause of road accidents and fatalities, especially distraction caused by the use of phones while driving. Current driver phone use detection methods can be divided into two broad categories, namely vision and non-vision-based approaches. Several methods need additional hardware infrastructure for the system to function. Approaches requiring little infrastructure will have improved adoption rates. The type of output produced by each method of implementation needs careful consideration. Some methods estimate phone position (i.e. is the phone located in the driver's or passenger's side of the vehicle), while other methods detect real-time instances of phone use (i.e. whether the driver is currently talking on the phone). Current vision-based methods can only detect a driver talking on the phone while it is held next to the ear. Arguably, even more dangerous phone use behaviour is texting, as it diverts a driver's attention for an extended period. This work focused on the implementation and combination of three driver phone use detection methods. Two of the methods provide phone localisation inside the vehicle. This is helpful, as it indicates if a phone might be in the driver's access area. The first localisation method utilises audio ranging to time the arrival of audio pulses; the second method uses embedded phone inertial sensors to track the phone from a known reference point. The third method monitors driver behaviour using a camera and identifies instances of phone use. This vision-based approach continually monitors the driver and detects talking on the phone and texting behaviour as it occurs. An approach that combines the phone use detection methods developed is proposed. Phone localisation methods are fused with driver behaviour image classification to create a more accurate and robust system. Comprehensive experimentation was conducted to test system performance in a wide variety of conditions and circumstances. Experiments were first conducted individually for each method; all methods were then tested together collectively. Collective evaluation involved numerous vehicle trips where factors such as harsh lighting conditions, head pose variations, increase in the ambient noise level and irregular phone pick-ups were tested. Experiments were chosen to evaluate method performance in real-world conditions. All experiments combined accounted for 122 minutes of collected data. This included a total of 7379 samples, 4839 samples were of 'no phone use', 1536 samples were of 'talking on phone' and 1004 samples were of 'texting'. Each sample relates to one second of recorded data. Results from experimentation show that very accurate localisation and driver behaviour identification can be provided by the methods developed. Audio ranging was the most accurate localisation method. It obtained overall average precision of 94.61% and recall of 96.22%. Phone inertial localisation achieved average precision of 83.50% and recall of 85.59%. The vision-based method that utilised a convolutional neural network (CNN) to classify driver behaviour yielded average precision of 91.47% and recall of 95.04%. CNN image classification combined with audio ranging localisation obtained even higher accuracy when detecting driver phone use behaviour. It obtained overall average precision of 95.89% and recall of 95.29% when classifying 'talking on phone' and 'texting' driver behaviour. A combination of detection methods increases system accuracy and robustness. A comparison of methods developed in this work to those in previous works illustrates that the new implementations provide several benefits and performance increases. The proposed solutions furthered the development of driver phone use detection systems. This contributes to the ultimate goal of lowering road accidents and fatalities caused by driver distraction. This work is based on the research supported wholly / in part by the National Research Foundation (NRF) of South Africa (Grant Number: UID 111723). This research was also supported by Telkom South Africa and Bytes Universal Systems via the Telkom Centre for Connected Intelligence (CCI) at the University of Pretoria. Electrical, Electronic and Computer Engineering MEng (Computer Engineering) Unrestricted 2020-02-20T09:51:57Z 2020-02-20T09:51:57Z 2020-04-29 2019 Dissertation * A2020 http://hdl.handle.net/2263/73454 en © 2019 University of Pretoria. All rights reserved. The copyright in this work vests in the University of Pretoria. No part of this work may be reproduced or transmitted in any form or by any means, without the prior written permission of the University of Pretoria. application/pdf University of Pretoria |
| spellingShingle | Phone use detection Phone localisation Driver distraction Behaviour classification Talking on the phone and texting detection UCTD Driver phone use detection using visual, audio and inertial sensor processing |
| title | Driver phone use detection using visual, audio and inertial sensor processing |
| title_full | Driver phone use detection using visual, audio and inertial sensor processing |
| title_fullStr | Driver phone use detection using visual, audio and inertial sensor processing |
| title_full_unstemmed | Driver phone use detection using visual, audio and inertial sensor processing |
| title_short | Driver phone use detection using visual, audio and inertial sensor processing |
| title_sort | driver phone use detection using visual audio and inertial sensor processing |
| topic | Phone use detection Phone localisation Driver distraction Behaviour classification Talking on the phone and texting detection UCTD |
| url | http://hdl.handle.net/2263/73454 |