Parking for Autonomous Vehicle using Deep Learning Approach

In this study, a model is proposed to allow the vehicle to park safely itself using deep learning approach for autonomous parking problem. The proposed method is included three main steps. In the first stage, the vehicle which a remotely controllable is designed to collect data. The goal of second part is prepared a track and label the collected data in order to collect real data with the designed vehicle. In an addition this collected data, is also added CNR parking dataset which is also based on real camera pictures. It has been used 75% of these data as training and 25% of them as testing. Lastly, a deep learning model has been developed based on CNN and transfer learning according to these data. This model classifies whether parking slot is empty or not. if classification is called free, then the vehicle starts to be parking maneuver, the vehicle doesn’t do parking maneuver while it is called busy. As a result, accuracy of the model was calculated about 0.96 which is quite good for park slot occupy detection. It was also performed successfully 7,6 out of every 10-parking maneuver.