الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 424 |  |  |
| | | from | 424 | | |
Abstract
In indoor navigation, cameras may be used as an aiding algorithm for inertial navigation. Also, fast and accurate image matching is an important task used in various applications in computer vision and visual odometry.
Recently many techniques for detection, description, and matching are available. A comparison was made between different algorithms offered by the OpenCV library to show which algorithm was the best and most robust against image distortions. The results showed that the ORB detector, the ORB descriptor, and either the BruteForce-Hamming or the BruteForce-HammingLUT matchers were favored to be used in indoor environments.
Moreover, the performances of some platform navigation solutions in indoor environments were assessed. The experimental results show that the stereo visual odometry technique is the most accurate method, but it should be aided with wall constraints to enhance navigation positional accuracies. This technique was used to form a features database. This features database contains several well-described keypoints whose 3-d world coordinates were known.
The main part of this thesis focuses on developing a simple and accurate navigation algorithm suitable for pedestrian people using smartphone sensors. The suggested pedestrian navigation solution was a dead reckoning solution based on the motion sensors aided by monocular visual odometry.
This suggested solution needs some preparation before starting the navigation. The testing area should be photographed using two cameras in order to apply the stereo visual odometry principle and form a features database for this area. Also, the plan of this testing area should be surveyed to be used when applying the wall constraints.
During the navigation, the position of the user will be determined based on the motion sensor measurements. When the solution drifted with time, the user should stop and capture an image using his camera. The corrected location of the user could be determined by comparing the captured image with the features database. Also, this correct location helps the user to estimate his corrected stride length. In addition, the wall constraints should be applied to enhance the navigation results. The accuracy of this suggested solution was about 1.7 meters with a closing error of about 0.3%.