Assisting drivers using parking predictions through an automobile app

Abstract

This research is part of SmartSkyAutomator project, partially funded by the University of Malta research fund. This dissertation will present research on three main areas of study; data science, pervasive computing and mobile applications. The parking problem at the University of Malta is taken as a use-case. The increase of vehicles is making traffic and parking increasingly unbearable in Malta, with delays on Maltese roads being almost triple the European average [1]. This dissertation aims to make the process of finding a parking space more efficient through a combination of data science and mobile applications.

Following an observation study, the model for the current study was created. A custombuilt drone was previously used in this project, however for this study a commercial one was used. The drone helped in building a dataset which was used to build a prediction model for parking based on historic patterns. The vehicle detection tool in [2] was used to evaluate the manual logs of the dataset obtaining satisfactory results, albeit presenting some limitations. Different Regression algorithms were tested on the dataset and the best one overall was selected for the prediction model. After considering various techniques, the decision was taken to build an automobile app using web technologies and a Node.js framework. Via this solution, predictions made on the dataset are stored in a MongoDB database, and passed to a driver through the application. A controlled experiment was designed which enabled the functional and usability testing of the application.

The controlled experiment served as an opportunity to test the functionality of the automobile application. Positive results from the usability study showed that participants found the app simple, effective and safe to use. The study found that drivers preferred using the buttons on the touch screen rather than voice commands to interact with the app. Drivers said they would recommend the app to drivers looking for a vacant parking space. The app achieved a very high overall Mean Rating Score and helped drivers in making decisions to find a parking space. Data was only collected at six different timeslots for the upper part of Car Park 6 at the University of Malta. The experiment would have been more realistic had a larger dataset been collected to include more timeslots and more parking facilities. Some of the limitations of the vehicle detection tool developed in [2] were addressed and the tool was used to evaluate the manual logs of the built dataset.