Online Detection and Tracking of 2D
Geometric Obstacles from LRF Data
Mateusz Przybyła
∗
Pozna´n University of Technology
mateusz.przybyla@put.poznan.pl
February 21, 2017
Abstract
This work proposes a method for detection and tracking of local geometrical obstacles from sequences of
two-dimensional range scans. Detected obstacles are represented with linear or circular models. Circular
obstacles are subject to tracking algorithm based on Kalman filter. Solutions to both the correspondence and
the update problem of the tracking system are provided. The occurence of obstacles fusion or fission is defined
and addressed. A by-product of the system is the information on tracked obstacles velocity. The algorithm is
dedicated to wheeled mobile robots with mounted planar laser range finders.
1. Introduction
E
nvironment
, in which the mobile
robots work, may be very irregular
and changing with time. Perception of
its geometry is one of the key features needed
to provide autonomy to mobile robots. With
the help of laser range finders (LRF), sensing
of surrounding material objects became simple
and reliable. However, using raw data pro-
vided by such devices is often not suitable for
well established algorithms of motion with ob-
stacle avoidance or path planning [
6
,
21
,
3
,
15
].
Extraction of more concise information from
acquired data is an essential step between sens-
ing and actuation.
There are two mainstream approaches of
spatial data representation: grid-based and
vector-based forms. Both of them can be mu-
tually used to represent the environment in
detail [
13
,
8
]. This work centres on extrac-
tion and tracking of 2D vector-based geomet-
ric objects, from data provided by horizontal-
∗
Faculty of Computing, Chair of Control and Systems
Engineering, ul. Piotrowo 3A, 60-965 Pozna ´n, Poland, (+48
61) 665-29-87
working LRFs. Such devices can provide only
local, spatial information due to their work-
ing principles (e.g. occlusion). Still, locally
detected objects can be exploited in such prob-
lems as: target detection and tracking, real-
time reactive obstacle avoidance, local path
planning, environment visualization, removal
of moving objects for SLAM.
The obstacles related to the operations per-
formed by mobile robots can have many forms.
The following list distinguishes several genres
of obstacles and provides examples:
Obstacles distinguished by geometry:
sparse (e.g. wire fence, foliage), dense
(e.g. wall, tree trunk), convex (e.g. furni-
ture, car), concave (e.g. hole in the ground,
cage).
Obstacles distinguished by color/opacity:
opaque (e.g. wooden doors, rock), reflec-
tive (e.g. mirror, polished metals), trans-
parent (e.g. glass doors, plexiglass wall).
Obstacles distinguished by corporeality:
material (e.g. solid box, textile curtain),
virtual (e.g. programmed restrictions, me-
chanical constraints).
1
