MOVING OBJECT’S POSE ACQUISITION FROM IMAGE SEQUENCE
Guozhong Su“ *, Jianging Zhang *, Shunyi Zheng?
e c > > J e
* School of Remote Sensing Information Engineering, Wuhan University, 129 Luoyu Road, Wuhan 430079, China
sgz gh(gsina.com.cn jqzhang@supresoft.com.cn syzheng(2)263.net
ter
KEY WORDS: Parapoint Feedback Iteration, Moving object, Pose measurement, RANSAC method, Image Sequence,
DLT (Direct Line Transform)
ABSTRACT:
Acquisition of pose parameters of moving objects is an important problem in many fields. This paper proposed a method to obtain
pose parameters of moving object (mainly aircraft) from image sequence under assumptions of monocular and perspective view as
well as known 3D shape and initial pose parameters of objects. Firstly image sequence is captured by optical-electric phototheodolite.
Based these images, a method for aircraft pose measurement from image sequence has been deeply studied from the view of
photogrammetry and projective geometry. A novel idea, Parapoint Feedback Iteration, has been presented and used in aircraft pose
measurement. Firstly, we derive nonlinear equations to estimate pose parameter with the correspondence between image feature and
model feature. To solve the nonlinear equations, at first several frames several control points must be selected manually and a DLT
(Direct Linear Transform) algorithm has been used to provide approximate value, in the following frames Kalman filter can be used
to predict motion parameter of next frame as approximate value. To get the accurate parameters, Parapoint Feedback Iteration
algorithm has been used. Several sets of data have been used to test the method and results show its reliability and robustness.
1 INTRODUCTION
[t is a hot topic to study and measure motion pose of object in
many fields. For example, motion pose of aircraft is important
to evaluate capabilities of aircraft control system and provide
key data for improvement of aircraft system design by analysis
of its space location. X( Y! Z! land movement pose ||, OK! !
at different times. In photogrammetry, known motion pose of
aircraft can be used to determine external parameters of aerial
photography taken by aircraft so as to reduce workloads of
locating GCPs (ground control point) in the field. Now
normally there are two methods to measuring aircraft pose: one
is to measure locations and poses of aircraft by use of GPS and
IMU loaded in aircraft, the other is high speed tracking and
recording moving objective by optical-electric phototheodolite,
and then identifying poses of moving objective with image data
and camera parameters. The former is called "interior
measurement of object pose", now its prevalent means is
identifying craft poses by using three high precision GPSs or
IMU and correcting signals received by GPS. The measuring
accuracy is restricted by GPS and IMU, as same as, flight
height and high speed changing pose of craft may cause GPS
signal to be lost, and IMU is very expensive and difficult to
operate. The later is called “exterior measurement of object
pose" and has some advantage over the former. Because such
method doesn't need to directly contact flying objectives and
can it be used widely. The paper mainly studies techniques and
means about “exterior measurement of motion object pose”.
Image sequence data obtained by optical-electric
phototheodolite have such characteristics: Imaging background
is simple, objective is single, geometry model data of object
can be known beforehand, the image data captured is huge.
Due to simple image background, aircraft objectives can be
*
Corresponding author.
extracted automatically from images sequences. However, due
to longer distance between aircraft and camera (3 Ui~10 LJ), size
of objective image is generally between about 100 and 1000
pixel, so that sometimes it is more difficult to determine {lying
pose of aircraft.
Now, there are many methods to measure objective pose by use
of image sequence. For example, Wenhao Feng calculated
objective pose by Direct Linear Transformation[l].
Horaud[7],etc proposed to calculate objective pose by four
point perspective means. DeMenthon,D[8],etc proposed to
calculate objective pose by a weak perspective camera mode.
Ohta, Y[2,3],etc put forward that model pose can be calculated
by parasperspective camera model.
However, it is necessary for methods mentioned above
to collect at least 3 to 4 GCPs for the calculation of objective
pose. In many cases it is very difficult to identity feature points
on images when aircraft is so far and aircraft is so small in the
image frame. To solve this question, we proposed a method of
matching simulation image and real image to trace objective
pose: First of all, the 3D model of aircraft is measured by use
of close-range photogrammetry method, then a simulation
system of optical-electric phototheodolite is set up which can
produce simulation image of aircraft in different pose, finally
aircraft pose at different times in image sequences are
calculated by matching real images and simulation images. In
this process, no GCP is needed and silhouette of aircraft is used
in calculation to determine the pose parameter of aircraft. It is
obvious that information in silhouette of aircraft is much more
abundant than that in several GCPs and therefore the results are
also much more reliable that can be seen in the experiment
below.
7049 ANY (UD
(IO —"C9 rm un
lae» 1
