Counting Cars and Determining the Vacancy of a Parking Lotusing Neural Networks

Sammanfattning: A lot of time, energy and money is being wasted when people are trying tond a parking lot. These elements could be reduced if the driver is provided vacancy information of a parking lot beforehand. In this thesis Google's Object Detection API is implemented and two pre-trained models are being used on the PKLot dataset to detect and count the number of cars in a parking lot. The models are based on a Region-based Convolutional Neural Network (R-CNN) which is explained in more detail. The models are compared with each other and its result presented. The result is presented with three factors in focus, the number of predictions made by the models, the number of cars a model missed to predict and how many objects that were wrongfully predicted. This was then tested on a Raspberry PI with the purpose to avoid using a remote computer for the image processing and prevent potential laws regarding camera surveillance. Finally, we determine if this functionality can actually be delivered using state-of-the-art technology.