Supervisors: Dr. Danda Pani Paudel, Prof. Angela Yao, and Prof. Luc Van Gool
In this thesis a crowdsourcing solution for the parking spot occupancy detection is proposed. The parking spot detection is done using only smartphones mounted inside vehicles. The cameras of the smartphones stream the video to a object detection algorithm to detect the parking spot occupancy status. The parking spot localization is done with manually marked parking spots within a 3D reconstruction of the area of interest. When the reconstructed area is being passed, the parking spots are projected into the image of the new camera. The pose of the new camera is found by solving the PNP problem. The object detection and parking spot detection part are running in real time on a smartphone. The 3D reconstruction and PNP problems are solved offline in advance. The final parking spot occupancy is displayed on a map. This thesis successfully shows that a parking spot occupancy detection with the described approach is possible. Empty parking spots are detected with a precision of 93%, recall of 85% and occupied parking spots are detected with a precision of 88% and recall of 79%. It is shown that it is possible to detect cars and the parking spot occupancy in real time on a smartphone. Therefore, similar approaches could be realised and ready for everyday use soon.