Performance Evaluation of Optimization Methods for Super- resolution Mosaicking on UAS Surveillance Videos

Unmanned Aircraft Systems (UAS) have been widely applied for reconnaissance and surveillance by exploiting the information collected from the digital imaging payload. However, the data analysis of UAS videos is frequently limited by motion blur; the frame-to-frame movement induced by aircraft roll, wind gusts, and less than ideal atmospheric conditions; and the noise inherent within the image sensors. Therefore, using super-resolution mosaicking of lowresolution UAS surveillance video frames, becomes a critical requirement for UAS video processing and important for further effective image understanding. In this paperwe develop a novel super-resolution framework which does not require the construction of sparse matrices. The proposed method applies image operators in the spatial domain and uses an iterated back-projection method to construct super-resolution mosaics from overlapping UAS surveillance video frames. The Steepest Descent method, Conjugate Gradient method and Levenberg-Marquardt algorithm are used to numerically solve the nonlinear optimization problem for estimating a super-resolution mosaic. A quantitative performance comparison in terms of computation time and visual quality of the super-resolution mosaic using the three numerical techniques is presented. The Levenberg-Marquardt algorithm provides a numerical solution for least squares curve fitting, which avoids the time-consuming computation of the inverse of the pseudo-Hessian matrix in regular singular value decomposition (SVD). The Levenberg-Marquardt method, interpolating between the Gauss–Newton algorithm (GNA) and the method of gradient descent, is efficient, robust, and easy to implement. The results obtained in our simulations shows a great improvement of the resolution of the low resolution mosaic of around 47 dB for synthetic images, and a considerable visual improvement in sharpness and visual details for real UAS surveillance frames. The convergence is rapid requiring typically ten iterations.