Kloosterman, H. (2007) Outdoor Slam Using Stereo Vision and SIFT Features. Master's Thesis / Essay, Artificial Intelligence.
|
Text
AI_Ma_2007_HKloosterman.CV.pdf - Published Version Download (3MB) | Preview |
Abstract
Simultaneous localization and mapping (SLAM) is a central subject in the field of autonomous robotics. SLAM is the process of localizing a moving robot in an unknown environment. This calls for localizing the robot relative to the map. On the other hand, the map has to be constructed while the robot moves, which makes SLAM a "chicken-and-egg" problem. Furthermore, the localization and the map must both be derived from sensor readings, which contain noise. In the frame of this study, a vision only SLAM implementation is achieved. Large-scale outdoor experiments are performed at TNO Defense, Security and Safety, in natural and urban terrain. The environment is perceived by sparse stereo vision using scale invariant image features (SIFT) as interest points. Successive stereo images are used to estimate the displacement of the robot, which causes a cumulative error. The localization error can be decreased when the robot is located at a point it has been before when the position of the robot was known more precise. Detecting such a point is called loop closure. To reduce the calculations needed for recognizing loops, feature selection can be used. For the experiment, the conclusion is that on a route of 1100 meters, by neglecting the features of the ground plane, the search space can be reduced by a factor two without notably degrading loop closure performance. Preselecting the landmarks on the map based on their location, reduces the number of landmarks to be matched towards the end of the run by a factor 50. Using absolute matching distances, compared to relative matching distances, ensures more constant loop recognition when the number of landmarks on the map grows. Storing, not all the landmark information multiple times on local stored maps, but only once, on a global stored map, strongly reduces computational power and memory usage.
| Item Type: | Thesis (Master's Thesis / Essay) |
|---|---|
| Degree programme: | Artificial Intelligence |
| Thesis type: | Master's Thesis / Essay |
| Language: | English |
| Date Deposited: | 15 Feb 2018 07:30 |
| Last Modified: | 15 Feb 2018 07:30 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/9071 |
Actions (login required)
 |
View Item |