Machine Vision for Locomotive Control

Wallin, Anthony (2021) Machine Vision for Locomotive Control. [USQ Project]

Text (Project)
WALLIN Anthony_dissertation_Redacted.pdf

Download (3MB) | Preview


Trains have existed since the 19th century. The locomotives and rolling stock have advanced greatly since the first iterations. The infrastructure that support these mighty machines has not kept up with the rapid rate of advancement. The purpose of this paper was to investigate a new method of automatic locomotive control using the latest in machine vision technology.

Automatic trains are very niche, used almost exclusively in light rail. The gap in automation between rail and other modes of transport steadily grows. Many things contribute to this, but a key factor is that the vast majority of infrastructure was build before the turn of the 21 st century with little consideration for new technologies. The latest automation uses short range wireless communication and complex servers to operate the trains. This makes such methods troublesome to implement into cross country freight networks.

The latest advancements in artificial intelligence and machine vision have brought algorithms with complex neural networks at the core. Many industries have already begun implementing such algorithms. Most prominently in the transportation industry, the technology has been implemented into cars to create an ‘autopilot’ feature. Tesla’s range of cars and soon trucks is the most successful implementation to date.

The algorithm that was tested for viability was YOLOv2. Six variations were created, tested then compared against each other. Data collected from a train driving simulator was used as training data and testing data. MATLAB was used to build and test these algorithms.

Results showed that the smaller algorithm that was created was superior in all cases. The results also proved that YOLOv2 based algorithms can identify rail signals. This has opened many paths for future work, from neural network and algorithm optimisation to taking the work done here and applying it to real locomotives.

Statistics for USQ ePrint 51848
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; Long, Derek
Qualification: Bachelor of Engineering (Honours) (Mechatronic)
Date Deposited: 03 Jan 2023 05:10
Last Modified: 26 Jun 2023 03:02
Uncontrolled Keywords: train, locomotive, automated control, machine vision, artificial intelligence, neural network, YOLOv2, algorithm

Actions (login required)

View Item Archive Repository Staff Only