Roman Števaňák, Adrián Matejov, Marek Šuppa, Ondrej Jariabka
Supervisor(s): Igor Farkas, Marek Nagy
Comenius University
Abstract: Parking space management is an issue that affects any building above certain size. While there are solutions which rely on networks of connected sensors for each parking space, these are usually difficult to maintain and expensive to implement. In this work we present PKSpace -- an opensource solution for detecting vacant and occupied parking spaces, using only an inexpensive camera and single-board computer. PKSpace provides the whole package: by being able to obtain the images, provide admin interfaces for specifying parking spaces, marking them as vacant or occupied, and training machine learning models, which are capable of predicting the number of vacant or occupied parking spaces in new images, it allows the user to deploy the system in a short amount of time while minimizing the amount of resources necessary for its maintenance. This work also includes a comparison of methods used for this task and their empirical evaluation on a newly introduced FMPH parking dataset.
Keywords: Computer Vision
Full text: Year: 2017
