Access Type

Open Access Thesis

Date of Award

January 2019

Degree Type


Degree Name



Electrical and Computer Engineering

First Advisor

Caisheng Wang

Second Advisor

Chinan Tan


Research and development in autonomous vehicles are currently very active since these vehicles are expected to play an important role in future transportation. During the past decade, numerous Advanced Driver Assistance Systems (ADAS) features have been developed and implemented on production vehicles, such as Adaptive Cruise Control and Lane Keeping Assist. However, most ADAS features are aimed to assist driving on highways where the environments are usually more structured and decision making can be made more easily. Non-highway environments, such as in the case of self-parking, are unstructured and requires more complicated analysis of the image information, localization and path planning. Most camera-based auto-parking features are limited to partial autonomy, not intelligent enough to park a vehicle automatically without the participation of a driver. Camera-based auto-parking systems can detect parking spots but cannot accurately localize the vehicle as well as the parking spot. The objective of this thesis is to design advanced sensor fusion and motion planning algorithms, especially for Automated Parking Assist (APA) systems. These algorithms are specifically designed for ADAS under low velocity and require high precision vehicle maneuvering, by increasing the localization accuracy, in terms of vehicle and parking spot location. The APA system will help the driver to search for available parking spots and drive the vehicle into the parking spot safely. The strategic architecture of the auto-parking system includes environment perception, sensor data fusion, motion planning, and vehicle control. Perception algorithms such as line detection and object detection are discussed in this thesis. Sensor data fusion and data association using extended Kalman filter for parking spot and vehicle location tracking are developed in this thesis. Rapidly-exploring Random Tree (RRT) motion planning algorithm is used to generate a path, leading the vehicle to park into the parking spot. Simulations for sensor data processing, data association and motion planning are conducted in a Robot Operating System (ROS) environment. A versatile virtual environment with vehicle dynamics model and control algorithms are developed in the simulation environment. Parking spot detection, vehicle behavior decision making, and motion planning are tested based on virtual sensor signals modelled in ROS. Simulation results show that the vehicle can drive automatically into the parking spot without the participation of a driver. Vehicle localization field experiments based on GPS sensor fusion has been conducted in an open parking lot. Localization accuracy of ego vehicle is improved.