Accuracy Analysis of a Dual Camera System with
an Asymmetric Photogrammetric Configuration
Bo Wu, Lei Ye, and Yuansheng Yang
Abstract
This paper presents a dual camera system combining a
wide field-of-view (
FOV
) surveillance camera and a pan-tilt-
zoom (
PTZ
) camera with an asymmetric photogrammetric
configuration, and focuses on the analysis of its attainable
measurement accuracy. First, we discuss the geometric
modeling of the asymmetric photogrammetric configura-
tion and analyze the accuracy of measurement based on
error propagation for different baseline lengths, different
focal lengths, and different pan angles of the
PTZ
camera.
Second, we performed a comprehensive accuracy analy-
sis based on Monte Carlo simulation, which incorporated
artificial noise into the input data. Third, we conducted
actual experiments in indoor and outdoor environments to
verify the theoretical and simulation results. We found that
the baseline length between the dual cameras was the main
factor influencing measurement accuracy. Increase of the
PTZ
camera focal length could improve the measurement
accuracy, but this trend was not significant when its focal
length was relatively long. The pan angle of the
PTZ
camera
also influenced the measurement accuracy, but this influence
was not significant at short ranges. From these discoveries,
we present an optimum configuration of the dual camera
system for better than 1 percent measurement accuracy of
the range within a normal observation range (e.g., 60 m). This
proposed dual camera system provides enhanced machine
vision capabilities that can be used in various applications.
Introduction
Traditional photogrammetric systems use stereoscopic image
pairs, obtained either through a pair of identical cameras with
a fixed lens rigidly connected by a hard base or through one
camera repeatedly imaging the interest area (Fraser, 1982; Li
et al
., 2004; Di and Peng, 2011). These systems are considered
symmetric photogrammetric systems, as the stereo cameras or
images are symmetrically distributed and the stereo images
have similar characteristics (e.g., image scale/resolution,
FOV
,
and coverage). Symmetric photogrammetric systems were
dominant in traditional photogrammetric applications in the
past; however, the intelligent vision capabilities of these sys-
tems are limited in non-traditional photogrammetric applica-
tions. For example, an ideal video surveillance system may re-
quire monitoring of a wide scene while zooming in on specific
targets of interest, and measurement and tracking of objects
using stereo images. These aims are impossible to achieve
with traditional, symmetrical photogrammetric systems.
We present a dual camera system with an asymmetric
photogrammetric configuration for better machine vision ca-
pability. The system combines a standard surveillance camera
and a
PTZ
camera, thus providing simultaneous wide
FOV
and
localized high-resolution imaging capability, in addition to
stereo measurement capability via the vertical base connecting
the dual cameras (see Plate 1). The asymmetry of the proposed
system is due to its asymmetric stereo configuration. The
surveillance camera has a short focal length but a large
FOV
,
whereas the
PTZ
camera has longer and varying focal lengths
but a small
FOV
. The surveillance camera is fixed, whereas the
PTZ
camera has adjustable pan and tilt angles. To render this
asymmetric dual camera system useful, the attainable mea-
surement accuracy of the system must be characterized.
Dual camera systems with asymmetric photogrammetric
configurations are not commonly used in photogrammetry.
Di and Peng (2011) investigated the wide-baseline mapping
capability of the Mars rovers in the 2003 Mars Exploration
Rover (
MER
) mission. Analysis of mapping accuracy with
respect to baseline length using the rover images was con-
ducted by theoretical derivation with error propagation, and
Monte Carlo simulation. However, all of the images used in
their research were acquired by the same camera. They also
assumed that the two camera axes were parallel to each other
and perpendicular to the baseline when collecting the images
in two different places, which simplified the derivation but
may not have represented the actual situation.
In the area of machine vision, there have been several
achievements in the use of dual camera or multi-camera sys-
tems for the purposes of 3
D
reconstruction (Luh and Klaasen,
1985; Allen and Bajcsy, 1986; Kanade
et al
., 1998; Gao
et al
.,
2000; Li
et al
., 2004). However, most of these studies have
used two identical cameras to generate the 3
D
information.
Marcenaro
et al
. (2002) proposed a multi-resolution outdoor
dual camera system, which contained a
PTZ
camera and a
wide-
FOV
camera and was able to recognize objects more ac-
curately than other systems. However, this system was only
used to investigate enhanced imaging and recognition, rather
than stereo measurement. Iwata
et al.
(2006) developed a
hybrid camera surveillance system for human tracking that
contained an omni-directional system comprising 36 cameras
for the acquisition of wide-range images, and a pair of
PTZ
cameras to acquire images of sufficient resolution for object
measurement and identification. Wan and Zhou (2009) also
proposed a stereo rectification method with dual symmetric
PTZ
cameras. Investigations into dual camera systems with
asymmetric photogrammetric configurations have been rare in
the past, and important aspects of asymmetric photogrammet-
ric configurations, such as theoretical accuracy analysis and
error propagation, have yet to be thoroughly investigated.
This paper focuses on the systematic accuracy analysis of
the dual camera system with an asymmetric photogrammetric
configuration. First, we present the detailed configuration of the
Department of Land Surveying and Geo-Informatics, The
Hong Kong Polytechnic University, Hung Hom, Kowloon,
Hong Kong (
).
Photogrammetric Engineering & Remote Sensing
Vol. 81, No. 3, March 2015, pp. 219–228.
0099-1112/15/813–219
© 2014 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.81.3.219
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
March 2015
219
