Probabilistic Semantic Classification of Trajectories

Introduction

The approaches described in previous chapters are able to classify static observations using a mobile robot. However, mobile robots are dynamic agents that move along different trajectories. When operating in indoor environments, robots usually have a moderate velocity and a relatively continuous movement. That means, that observations obtained by a mobile robot at nearby poses are typically very similar. Furthermore, certain transitions between classes in a trajectory are rather unlikely. For example, if the classification of the current pose is kitchen, then it is rather unlikely that the classification of the next pose is office given the robot moved a short distance only. To get from the kitchen to the office, the robot first has to move through a doorway.

The work presented in this chapter originated from a collaboration with Axel Rottmann.