Relative orientation estimation of video streams from a single pan-tilt-zoom camera

초록

As a surveillance camera for the security of urban area, a single pan-tilt-zoom (PTZ) camera has been used. For effective application of the video sequence of the single PTZ camera, accurate orientation or calibration of the video sequence is needed. This paper proposes a mathematical model based on the collinearity condition equation for relative orientation estimation between frames from video sequence of a single PTZ camera. The basic concept of this mathematical model is similar to the homography concept, which explains the relationship between two images. However the proposed model is based on the collinearity equation and the model can estimate the varying focal length and relative rotation angles directly from corresponding points between image frames. For the relative orientation estimation by the model, the ratio between focal length and CCD size of the first frame is required as initial input. In this paper, we estimated the ratio by two vanishing points. The corresponding points were extracted by Scale Invariant Feature Transform (SIFT) and RANdom SAmple Consensus (RANSAC) automatically. Photos taken by digital camera and a video sequence of a single PTZ camera were used for experiments. The experiment results show the proposed method can estimate the relative orientation from the video sequence automatically.