Movement estimation of a robot using stereo vision

In this work we focus solely on the problem of localisation, comparing different estimation algorithms of the trajectory taken by a robot from the observations and readings obtained by the robot itself. In our problem, we will work with images taken by a stereoscopic vision system of uncalibrated cameras, we will assume that the movement of the robot is on a flat surface and we will use natural landmarks. As we will see, the information obtained from this type of sensor allows a robust estimation of movement taken between each pair of observations without the need to use the information from the robot's propioceptive sensors. The solution of this problem, known as visual odometry, is critical within the majority of subsequent navigation processes.