Assistive Parking Systems Knowledge Transfer to End-to-End Deep Learning for Autonomous Parking

In numerous spots, parking a vehicle is challenging task and requires an experienced driver to maneuver and park the vehicle efficiently. With the advent of Automatic Parking Assist Systems (APAS), drivers can park their vehicles automatically and safely. These systems, however, still require driver intervention and constant attention while parking. The APAS system uses the onboard sensors to perceive the environment to identify the obstacles around and a proper parking space. The system then plans a collision-free trajectory and follows that trajectory to park the vehicle in the designated parking space. This paper presents an intelligent parking system for parking Unmanned Ground Vehicle (UGV) perpendicularly using Convolution Neural Networks (CNNs). To overcome the problem of dataset scarcity and quality APAS system is used to generate training data. The neural network model is trained to mimic the APAS system behavior captured in the generated dataset. The evaluation of the trained CNN model showed that the proposed intelligent parking system is able to park the vehicle perpendicularly with accurate orientation.