Abstract:
|
The design and development of intelligent service mobile robots that interact
with humans in daily living activities or cooperate with persons in specific
tasks, requires the design of new tools that allow the understanding of human
motion intentionality. We can find examples of this kind of requirements for
guiding people, for meeting persons whether in indoor or outdoor environ-
ments or for doing robust navigation in highly crowded spaces. In any case,
the knowledge of human motion intentionality might allow to optimize the
trajectory described by the robot towards a more harmonized interaction
in environments typically inhabited by humans and help to find the best
human-robot motion behavior.
Tackling the problem by only relying on a robust navigation is not enough,
although this topic has grown enormously in the few past years. Thus, the
understanding of human motion in urban environments is of extreme impor-
tance in order to adapt service robots to typical human environments and
not in the contrary.
The present Master Thesis is a study in depth of the most relevant tech-
niques that tackle the problem of Human Motion Prediction (HMP). Specif-
ically, this document is focused on solving the HMP in outdoor and free
spaces.
To this end, it has been developed a geometrical method for generating
optimal trajectories based on the concatenation of cubic polynomials using
the so-called Collocation method. The main drawback of the aforementioned
method remains in its initialization. Regarding this issue, an initial estimated
solution has also been proposed to accelerate the calculation and to guarantee
convergence to a feasible solution. Results show a robust method for human
motion estimation in addition to a good calculation performance, making the
proposed algorithm a valid option to consider in real-time applications.
Furthermore, a statistical study of a large amount of trajectories was
done aiming to obtain a simple and accurate model of the basics for human
motion. Once obtained a model, concretely in the shape of a probability
density function, we can build a method for predicting using probabilistic
methods.
As a validation, both prediction methods are evaluated in a real outdoor
scenario through the Edinburgh pedestrian data base [14]. |