Estudiante

Walking Pedestrian Recognition
Crist´bal Curio, Johann Edelbrunner, Thomas Kalinke, Christos Tzomakas, and Werner von Seelen
o
Institut f¨r Neuroinformatik,
u
Lehrstuhl f¨r Theoretische Biologie
u
Ruhr-Universit¨t Bochum, 44780 Bochum, Germany
a

Contact Author:

Crist´bal Curio
o

Address:

Ruhr-Universit¨t Bochum
a
Institut f¨r Neuroinformatik
u
44780 Bochum
GermanyTelephone:

+49-234-322-4231

Fax:

+49-234-321-4209

Email:

Cristobal.Curio@neuroinformatik.ruhr-uni-bochum.de

Walking Pedestrian Recognition
Crist´bal Curio, Johann Edelbrunner, Thomas Kalinke, Christos Tzomakas, and Werner von Seelen.
o
Abstract — In recent years many methods providing the
ability to recognize rigid obstacles - sedans and trucks - have
been developed. Thesemethods provide the driver with relevant information. They are able to cope reliably with scenarios on motorways. Nevertheless, not much attention has
been given to image processing approaches to increase the
safety of pedestrians in urban environments. In this paper
a method for the detection, tracking, and final recognition
of pedestrians crossing the moving oberserver’s trajectory
issuggested. A combination of data- and model-driven approaches is realized. The initial detection process is based on
a fusion of texture analysis, model-based grouping of, most
likely, the geometric features of pedestrians, and inverseperspective mapping (binocular vision). Additionally, motion patterns of limb movements are analyzed to determine
initial object hypotheses. The tracking of thequasi-rigid
part of the body is performed by different algorithms that
have been successfully employed for the tracking of sedans,
trucks, motorbikes, and pedestrians. The final classification
is obtained by a temporal analysis of the walking process.

in the near field. Another stereo and intensity based system to detect pedestrians and basic actions in static background scenes is described in [4].Here, a new method for walking pedestrian recognition is
presented. It is characterized by a successive processing at
various levels of description and their integration yields the
final recognition. The method is restricted to the detection
of pedestrians that cross the road (see Figure 1).

Keywords — image analysis, inverse perspective mapping,
Hausdorff-Distance, texture, spatio-temporalpattern, data
fusion, system architecture

I. Introduction

I

N recent research many approaches for recognizing human shapes in indoor and outdoor scenes have been developed. Different sensors have been used, such as thermal
sensors and CCD cameras. Although thermal sensors can
yield accurate results, they defunct as soon as the environment’s temperature reaches a certain level so thatobjects cannot be reliably separated from the background.
In vision-based approaches human movements can be detected by the subtraction of subsequent image frames as
long as the camera does not move and the lighting conditions change slowly. In the case of a moving observer, the
ego-motion implies additional motion in the background
making the detection of independent motion a nontrivial
task. Ingeneral, there are no common features for the
recognition of pedestrians. In [1] a large number of features has been used to build up a neural classifier. Still,
the number of features is too high to ensure fast computation and detection has mainly been restricted to rear and
front views of pedestrians. In another approach [2] a timedelay neural network was trained to recognize pedestrians
fromspatio-temporal receptive fields. In that approach the
initial detection is based on a stereo camera system. Stereo
vision is useful for short and middle distance ranges. The
classification is done by analyzing the raw intensity values
where background texture influences the signal to noise ratio. Another system has been introduced which relies on an
initial stereo segmentation step and...