Recursive total least squares algorithm for 3-D camera motion estimation from image sequences

We present the estimation method of global motion parameters corresponding to 3-D camera motion in the non-stationary noisy situation. The total least squares problem is first formulated to represent the global motion parameters estimation procedure from the noise-corrupted image coordinates. Then, a recursive total least squares (RTLS) algorithm is proposed to estimate 3D camera motion parameters in image sequences. The algorithm is proposed based on a five camera parameter model: zoom, focal length, pan, tilt, and swing. In the experimental results, the efficiency of the proposed RTLS algorithm is shown by comparing its MSE and PSNR with those of the conventional linear algorithms.