Abstract
This thesis describes an algorithm for 3D pose
calculation of known
objects from a single perspective view. The pose
calculation algorithm
consists of four main stages: pose candidate generation,
candidate
verification, optimization and reliability analysis. In
the first stage, pose
candidates are generated by matching model and image
triangles. An
important subproblem, the socalled 3point perspective
pose estimation
problem, is solved numerically. Pose candidates are
verified by matching
model and image features. The first accepted candidate is
then optimized.
Three optimization methods are considered:
NewtonRaphson, simulated
annealing and robust Mestimate using Lorenzian
distribution. The reliability estimates are based on
covariance analysis.
Uncertainties due to
both model and image noise are taken into account. A
structure
optimization method to reduce modeling noise is also
presented.
The algorithm was implemented on two parallel computers,
a Connection
Machine2 (CM2) and a Hathi2/16. The Connection Machine
is an SIMD
machine whereas the Hathi2/16 is based on the MIMD
architecture. To
evaluate the speed and accuracy of the algorithm, several
experiments were
performed using both synthetic and real images. In
simulations, the
accuracy of the calculated pose, the validity of
uncertainty estimates and
the effect of dynamic modeling were studied using a
polyhedral test object
whose maximum distance between any two vertices was 175
mm.
Simulations showed that for the distance range from 0.5 m
to 2.75 m, the
standard deviation of the rotation error ranged from 0.5
degrees to 3.5
degrees when the standard deviation of image and model
noise was 1.5
pixels and 1.5 mm, respectively. For the xand
ydirections the errors
ranged from less than 2 mm up to 10 mm, and for the
zdirection, from a
few millimeters up to 75 mm. Much better results were
achieved at the
lower noise levels. The processing times on both the
machines were
approximately the same. On the Hathi2/16, when all the
16 processors
were used, the pose calculation time ranged from 0.376 s
up to 0.814 s.
For ease of programming the Hathi2/16 was better than
the CM2, and it
was also considerably cheaper.
Original language  English 

Qualification  Doctor Degree 
Awarding Institution 

Supervisors/Advisors 

Award date  18 Dec 1992 
Place of Publication  Espoo 
Publisher  
Print ISBNs  9513842401 
Publication status  Published  1992 
MoE publication type  G4 Doctoral dissertation (monograph) 
Keywords
 3d analysis
 calculation
 parallel computing
 modelling
Fingerprint Dive into the research topics of 'Calculation of 3D pose of a known object in a single perspective view: Dissertation'. Together they form a unique fingerprint.
Cite this
Pehkonen, K. (1992). Calculation of 3D pose of a known object in a single perspective view: Dissertation. VTT Technical Research Centre of Finland.