POSE DETERMINATION FROM A SINGLE IMAGE
IN A CONTROLLED CAD ENVIRONMENT
|
A. D. Styliadis The ATEI of Thessaloniki Dept. of Informatics Agiou Panteleimona 24 55133, Thessaloniki, GR |
C. J. Karagiannidis The ATEI of Thessaloniki Dept. of Informatics Agiou Panteleimona 24 55133, Thessaloniki, GR |
N. C. Zestas The Aristotle University of Thessaloniki P.O. Box 469 54124, Thessaloniki, GR nzestas@topo.auth.gr |
K. N.
Chatzara Synthesis Interactive Katouni 28 554 25, Thessaloniki, Greece synthe@otenet.gr |
ABSTRACT
The purpose of this work is to derive an easy-to-use interactive method, using simple CAD software, for sensor attitude (camera pose determination, i.e. camera positioning (Xo, Yo, Zo) and orientation (ω, φ, κ rotations)) and object pose determination, with a 10-3 accuracy, from a single image. It is actually a control-point-free method intended for terrestrial photogrammetric engineering applications (mainly in Architecture and Archaeology dealing with facades) and for robotics applications (mainly in robot location-tracking and object recognition).
This new method is based on mutually parallel or perpendicular line pairs in observed rectangular shape images usually found in photography. In man-made environments, rectangular shapes can be seen everywhere. It is thus convenient to use rectangular shapes for pose and object determination in photogrammetric engineering (close-range space rejection) and robotics (robot location).
The proposed system is to some degree automatic, since it analytically computes discrepancy vectors and examines possible solutions to minimize a penalty function. These different solutions are animated on the screen and the procedure is simply stopped by the user when he feels the solution was achieved. When the solution deviates user interaction is possible. As a supplement to automatic procedure a manual approach is provided for micro-corrections and enhancements. This manual approach is based on mouse movements and appropriate push-button selections according to a predefined step.
Some merits of the proposed method are: Single photography of known rectangular shapes is sufficient. Multiple rectangular shapes can be utilized to promote the accuracy. Partial rectangular shapes, usually found in old photography, also can be utilized. Finally, the major merit of the method is that the solution can be uniquely determined in an interactive user-friendly CAD environment.
Experimental results show the feasibility of the proposed approach. Accuracy evaluation for determining pose determination precision is also included.
Keywords: CAD programming, image understanding, camera calibration, pose determination, virtual camera.