ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., I-3, 353-358, 2012
© Author(s) 2012. This work is distributed under
the Creative Commons Attribution 3.0 License.
23 Jul 2012
R. A. Persad, C. Armenakis, and G. Sohn Geomatics Engineering, GeoICT Lab Earth and Space Science and Engineering York University 4700 Keele St., Toronto, Ontario, M3J 1P3 Canada
Keywords: Image sequences, CAD wireframe, Matching, PTZ camera, Orientation, LR-RANSAC, 3D localization Abstract. The tracking of moving objects from single images has received widespread attention in photogrammetric computer vision and considered to be at a state of maturity. This paper presents a model-driven solution for localizing moving objects detected from monocular, rotating and zooming video images in a 3D reference frame. To realize such a system, the recovery of 2D to 3D projection parameters is essential. Automatic estimation of these parameters is critical, particularly for pan-tilt-zoom (PTZ) surveillance cameras where parameters change spontaneously upon camera motion. In this work, an algorithm for automated parameter retrieval is proposed. This is achieved by matching linear features between incoming images from video sequences and simple geometric 3D CAD wireframe models of man-made structures. The feature matching schema uses a hypothesis-verify optimization framework referred to as LR-RANSAC. This novel method improves the computational efficiency of the matching process in comparison to the standard RANSAC robust estimator. To demonstrate the applicability and performance of the method, experiments have been performed on indoor and outdoor image sequences under varying conditions with lighting changes and occlusions. Reliability of the matching algorithm has been analyzed by comparing the automatically determined camera parameters with ground truth (GT). Dependability of the retrieved parameters for 3D localization has also been assessed by comparing the difference between 3D positions of moving image objects estimated using the LR-RANSAC-derived parameters and those computed using GT parameters.
Conference paper (PDF, 1595 KB)

Citation: Persad, R. A., Armenakis, C., and Sohn, G.: INTEGRATION OF VIDEO IMAGES AND CAD WIREFRAMES FOR 3D OBJECT LOCALIZATION, ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., I-3, 353-358,, 2012.

BibTeX EndNote Reference Manager XML