This project is a parking lot detection in order to automate the detection of the number of cars parked in a parking space. By using a pre-trained VGG16-net in combination with three fully connected layers we achieve an accuracy of 97% on overall. In this research we also investigate the influence of training on different parking lots and also considering different weather scenarios. Finally we look into ways to make the system of car detection truly automatic by automating the manual pre-processing step of parking lot marking.