Computer vision-based generating and updating of the public transit bus stop inventories

An updated asset inventory enables public transit agencies to make informed decisions on the maintenance and improvement of their physical assets. Conventional asset inventory surveys mainly rely on manual site visits and subsequent analysis, which are time consuming and expensive. Many research projects developed methods to automate condition assessment of civil infrastructure assets, such as road surfaces, structures, and sewage systems; however, research on the automated detection and condition assessment of public transit infrastructure is very limited. This research aims to contribute to addressing this gap by introducing an automated computer vision-based system to detect main assets in transit bus stops and update asset inventories using video frames captured by on-board cameras on operating buses. This system uses existing hardware systems on public buses to gather required data and then uses Deep Convolutional Neural Networks (DCNNs) to recognize public transit assets. In addition, a related method was proposed to process manually collected images for semi-automated asset inventory updating. The experimental results showed more than 95% detection rates in videos, which demonstrate potentials for practical applications.