Monitoring Dangerous Goods on Roads Using Computer Vision
Puumala, Joram (2023)
2023
Tietojenkäsittelyopin maisteriohjelma - Master's Programme in Computer Science
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2023-05-08
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202303102877
https://urn.fi/URN:NBN:fi:tuni-202303102877
Tiivistelmä
The transportation of dangerous goods on roads poses significant risks, as accidents involving hazardous materials can be deadly and devastating, especially in densely populated areas and confined spaces like tunnels. In the event of an accident, fast emergency response is crucial. Vehicles carrying dangerous goods must be marked with orange hazmat plates, one attached to the front and another to the back of the vehicle. This establishes a universal way to recognize these trucks within ADR member states. This thesis aims to determine if it is possible to use computer vision to automatically monitor dangerous goods in an efficient manner based on the visual information provided by the hazmat plates.
This research began with a review of relevant literature and background information. Then, a dataset was created for training deep learning-based object detectors to identify and locate trucks and hazmat plates in images. A method for classifying trucks as carrying dangerous goods was also developed. Finally, the effectiveness of the method was tested on a set of videos of highway traffic that were collected manually.
The proposed approach for dangerous goods detection and monitoring in this thesis proved to be highly accurate. The method was evaluated using a set of highway traffic videos with varying levels of complexity, including scenarios where false positives could occur. The proposed method was able to identify all trucks carrying dangerous goods and avoid false positives. The method also proved to be highly efficient, as it could process 26 frames per second on a modern edge device.
This research began with a review of relevant literature and background information. Then, a dataset was created for training deep learning-based object detectors to identify and locate trucks and hazmat plates in images. A method for classifying trucks as carrying dangerous goods was also developed. Finally, the effectiveness of the method was tested on a set of videos of highway traffic that were collected manually.
The proposed approach for dangerous goods detection and monitoring in this thesis proved to be highly accurate. The method was evaluated using a set of highway traffic videos with varying levels of complexity, including scenarios where false positives could occur. The proposed method was able to identify all trucks carrying dangerous goods and avoid false positives. The method also proved to be highly efficient, as it could process 26 frames per second on a modern edge device.