 |
Stephen Thomas |
   |
| Project:
|
| |
| Supervisor:
|
| |
| Host University:
|
| |
| Personal introduction |
top |
|
Hello,
I am from Waitakere City in the west of Auckland, New Zealand. I am a highly motivated, innovative, experienced graduate who is always seeking new challenges. My first-rate academic grades and high-level work experience reflect my ability to learn quickly and work well in both indiviudal and team environments. I have an excellent background in computer vision, embedded systems, telecommunications, robotics and control systems but my interests extend far beyond these fields. I consider my problem solving skills my biggest asset and am always searching for interesting interdisciplinary projects which enable me to expand my knowledge further. Besides my research interests I have a strong passion for the outdoors and enjoy sailing, surfing, tramping and kayaking as well as playing football both socially and competitively.
|
|
| |
|
Simultaneous localisation and mapping (SLAM) has been the focus of intensive research in the last decade due to the potential benefits it offers to the field of autonomous mobile robotics. SLAM is concerned with the ability of an autonomous vehicle to navigate through an unexplored environment and incrementally construct a map of the environment and localise itself within this map. This thesis describes an entirely vision-based, large-area, 6DoF SLAM system that was developed specically for real-time deployment on an autonomous underwater vehicle (AUV) equipped with a calibrated stereo system.
|
|
| |
|
This thesis represents an important contribution to entirely vision-based 6DoF SLAM as very few implementations currently exist, and the approach utilised in this thesis achieves comparable results and has the potential to operate in real-time. This SLAM system is based on the extended Kalman filter (EKF) and incorporates a novel approach to landmark description and data association in which landmarks are essentially local submaps that consist of a cloud of 3D points and their associated SIFT or SURF descriptors. Furthermore, landmarks are sparsely distributed in the constructed map which greatly simplifies and accelerates data association and map updates. In addition to performing localisation based on landmark observations the system also performs visual odometry and predicts vehicle motion using a constant-velocity model. By applying epipolar constraints to identify outliers the system is able to operate when Gaussian noise and outliers are present in the observations. The method has been verified in a simulated, texture-mapped 3D environment in which ray tracing is used to generate synthetic images for the two cameras at each vehicle position. |
|
| |
| I would like to thank my supervisors, Dr. Yvan Petillot and Dr. Joaquim Salvi as well as all the vibot staff and administration for their hard work and support during the 2 years of the Master. In addition, I would like to express my thanks to my family, girlfriend and all my friends (including the 26 vibotians) who have made this experience so enjoyable! |
|
| |
| MSc Thesis Project documents |
top |
|
| |
