3D Intruder Detection System with Uncalibrated Multiple Cameras

Satoshi KAWABATA  Sinsaku HIURA  Kosuke SATO  

D - Abstracts of IEICE TRANSACTIONS on Information and Systems (Japanese Edition)   Vol.J91-D   No.1   pp.110-119
Publication Date: 2008/01/01
Online ISSN: 1881-0225
Print ISSN: 1880-4535
Type of Manuscript: PAPER
intruder detection,  projective reconstruction,  visual hull method,  planer perspective projection,  convex hull,  

Full Text(in Japanese): PDF(1.5MB)
>>Buy this Article

In this paper we propose a practical intrusion detection system using uncalibrated multiple cameras. The introduced detection algorithm is based on a contour-based multi-planar visual hull method and a projective reconstruction method. As preparation for detection, user can use a marker to input corresponding points to define a restricted area in the scene instead of technical calibration. Then, the system reconstructs the coordinates of the points in the projective 3D space, and computes a convex polyhedron as the restricted area using the quick hull method, where covers the whole input points. As executing of detection, the system transforms each contour vector of object silhouette on each camera plane onto a sensitive plane, and calculates the common region on the sensitive plane as the intruding region. Because when an intruding object exceeds one of sensitive planes, the contours on the plane must have some the common regions in accordance with the visual hull method. Although, there is no need to reconstruct the whole shape of the object in 3D for detection, but the system can handle an arbitrarily 3D volumetric intrusion area by multiple planer perspective projections. The advantage of the system is that the common intricate calibration process for multiple camera system is completely veiled by the projective reconstruction technique as "detection volume setting" from users' view.