Real-time motion tracking of mobile robots via image registgration

• Main Author: Soroushi, Atousa
• Format: Others
• Language: English
• Published: 2009
• Online Access: http://hdl.handle.net/2429/6040

In order to have accurate, safe and reliable remote control for mobile robots, it is necessary to track their motion. With new high technology systems, fools for imaging and image acquisition are becoming less expensive every day. Consequently, vision-based systems are considered practical for tracking purposes. Most of the present motion-detection systems used for control of heavy duty hydraulic machines depend on external equipment like inertial sensors, deadreckoning or laser beacons. There are also a few existing systems for navigation purposes that are based on vision, but they do not perform in real-time or they require a partially known environment. The goal of this thesis is the design and implementation of a stand-alone motiontracking system, to track the local motion of a machine-mounted down-looking camera in realtime. By installing the camera under the base of a mobile robot, the local trajectory of that robot in an unstructured environment can be tracked. In the present work we use a computational vision approach to design an image registration system to estimate the two dimensional translation, rotation and small scaling factor from two partially overlapping Images. This is done in real-time (about 20 frames per second) even when there is simultaneous rotation and translation occurring between the two successive frames and images may have few significant features. To reduce the processing time and obtain information from different regions of the image, we process only selected regions of interest of the original image. An initial estimate is used to select those regions, so that regions contain similar features with the least possible disparity. This method, with the aid of a coarse-to-fine search method, makes the motion-detection algorithm able to execute in about 45ms and detect a dependable rotation of up to 11.25 degrees and translations of up to 8 pixels for each pair of image frames. === Applied Science, Faculty of === Electrical and Computer Engineering, Department of === Graduate