Moving Object Detection from Mobile Platforms Using Stereo Data Registration - Computational Intelligence Paradigms in Advanced Pattern Classification - page 25

Chapter 3
Moving Object Detection from Mobile
Platforms Using Stereo Data Registration

Angel D. Sappa 1 ,DavidGeronimo 1 , 2 , Fadi Dornaika 3 , 4 ,
Mohammad Rouhani 1 , and Antonio M. Lopez 1 , 2

1

Computer Vision Center
Universitat Autonoma de Barcelona, 08193, Bellaterra, Barcelona, Spain

2

Computer Science Department
Universitat Autonoma de Barcelona, 08193, Bellaterra, Barcelona, Spain

3

University of the Basque Country, San Sebastian, Spain

4

IKERBASQUE, Basque Foundation for Science, Bilbao, Spain
{asappa,dgeronimo,rouhani,antonio}@cvc.uab.es ,
fadi [e-mail]

Abstract. This chapter describes a robust approach for detecting moving objects
from on-board stereo vision systems. It relies on a feature point quaternion-based
registration, which avoids common problems that appear when computationally ex-
pensive iterative-based algorithms are used on dynamic environments. The proposed
approach consists of three main stages. Initially, feature points are extracted and
tracked through consecutive 2D frames. Then, a RANSAC based approach is used
for registering two point sets, with known correspondences in the 3D space. The
computed 3D rigid displacement is used to map two consecutive 3D point clouds
into the same coordinate system by means of the quaternion method. Finally, mov-
ing objects correspond to those areas with large 3D registration errors. Experimental
results show the viability of the proposed approach to detect moving objects like ve-
hicles or pedestrians in different urban scenarios.

1 Introduction

The detection of moving objects in dynamic environments is generally tackled by
first modelling the background. Then, foreground objects are directly obtained by
performing an image subtraction (e.g., [14], [15], [32]). An extensive survey on mo-
tion detection algorithms can be found in [21]. In general, most of the approaches
assume stationary cameras, which means all frames are registered in the same co-
ordinate system. However, when the camera moves, the problem becomes intricate
since it is unfeasible to have a unique background model. In such a case, moving
object detection is generally tackled by compensating the camera motion so that all
frames from a given video sequence, obtained from a moving camera/platform, are
referred to the same reference system (e.g., [7], [27]).
Moving object detection from a moving camera is a challenging problem in com-
puter vision, having a number of applications in different domains: mobile robots