Real-Time Pan-Tilt-Zoom Camera Calibration in Virtual Scene

Pan-tilt-zoom (PTZ) cameras are pervasive in Modern Society. However, research on the mixed reality system of PTZ Cameras is scarce due to the flexibility of PTZ cameras relying on visual information without sensors feedback and the complexity of the real world, such as lighting changes, occlusion, and lack of texture. To solve this problem, we propose a motion estimation model for PTZ cameras obtained from real PTZ camera motion observation. We show how the motion estimation model real-time estimates the pose of the camera in moving. Next, we propose a state-of-art PTZ calibration algorithm based on key-rays collection. We first extract SIFT features from the scene of the PTZ camera to build a key-rays collection. Then we obtain a group of the parameters of the current image against the collection using the two-point method and find best the parameters by the idea of RANSAC. In different real scenes, we evaluate the motion estimation model and calibration method on different cameras, demonstrating the accuracy of the motion estimation model and the effectiveness of the calibration.