Calibration of a Ptz Surveillance Camera Using 3d Indoor Model

Security monitoring, event detection and tracking of people are a few of the well-known applications for indoor video surveillance systems, but mainly they are accomplished by visual inspection of the images. There is need to integrate visual surveillance systems with 3D geospatial data to semantically improve interpretation of scene dynamics which is challenge in 2D image analysis. In this study a single view video camera is calibrated with respect to a 3D wireframe model to lay the foundation for further research into people tracking within this 3D space. For the estimation of camera parameters, the proposed method caters for the motions of a typical Pan-Tilt-Zoom (PTZ) surveillance camera. The method takes into consideration the degrees of freedom of the PTZ camera where the perspective centre is considered fixed –no translation change between the image framesand that the camera rotates about its centre. For the initial estimation of camera parameters we use point correspondence between image to model space followed by line correspondences within image space for updating the parameters after camera motion. Point correspondences are utilized in a self-calibration adjustment to project the 3D model an initial image frame. For the determination of the camera parameters of another overlapping image frame 2, the 3D model is projected into it using the camera parameters of the initial frame. Extracted lines are labelled on image 2 and a distance function is used to describe the relation of camera parameters between the two image frames based on the difference of the orthogonal distances between extracted lines and the projected model lines. Using image and model line correspondence, the distance function is minimized and the initial camera parameters are updated to retrieve those of the second image. The distance minimization approach for estimating camera parameters is tested on simulated and real datasets providing promising preliminary results.