A self-calibration method for smart video cameras

The automatic calibration of the intrinsic camera parameters such as the focal length and the camera orientation is an important pre-requisite for many computer vision algorithms in video surveillance. Despite its importance only a few number of methods show their applicability in embedded systems. This paper shows new results of previous work done in camera self-calibration on images of the York Urban data-set. These 102 images show typical urban scenes one might expect in practice. The evaluation shows that in 52 of 102 images the proposed method achieves less than 5% relative error in the focal length at a mean computation time per image of 14.45 s on a standard PC. We believe that these results show a fair balance between accuracy and computational performance and encourage an embedded implementation on a smart camera.