Real-time people tracking for mobile robots using thermal vision


  Real-time people tracking.pdf (Size: 2.13 MB / Downloads: 1)


Abstract

This paper presents a vision-based approach for tracking people on a mobile robot using thermal images. The approach combines a particle
filter with two alternative measurement models that are suitable for real-time tracking. With this approach a person can be detected independently
from current light conditions and in situations where no skin colour is visible.

Introduction

As research robots move closer towards real applications
in populated environments, these systems will require robust
algorithms for tracking people to ensure safe human–robot
cohabitation. The tracking problem is well-known in computer
vision, where there are developed systems used mostly in
surveillance and identity verification. Traditionally, tracking
multiple people has been investigated for surveillance
applications with one or more static cameras (see e.g. [16]) and
in automatic motion capturing and human motion analysis [24,
27]. In these environments, moving objects can be easily
detected using background subtraction techniques.

Related work

Feyrer and Zell [11] combine skin colour, shape, motion
and depth information to track a single person that faces the
robot. This approach is combined with information from a laser
scanner by the same authors in [12]. Schlegel et al. [30] extract
a colour histogram from the shirt of a person and combine
colour blob tracking with a contour-based approach. Tracking
of the head–shoulder contour and the hand silhouette is used
in [23] to build a human–robot interface. Colour histograms,
optical flow and an extension of the background subtraction
technique is used by Zajdel et al. [38] who developed a system
that keeps track of humans who leave the field of view of
the robot and re-enter.

Evaluation

This section presents the evaluation methodology used in
our experiments. First we describe the format and process of
acquiring ground truth data, the training procedure, then the
metrics used for evaluation and finally experimental results.

Conclusions and future work

This paper presented a people tracking system for mobile
robots using thermal vision and provided a thorough evaluation
of its performance. The system uses a robust and fast
tracking method based on a particle filter and several different
measurement models that are very fast to calculate (contourbased,
integral image features and a combination of the
two). We determined the optimal values of different system
parameters.