A GPU accelerated algorithm for compressive sensing based video super-resolution

This paper presents a parallel algorithm designed for video reconstruction based on Compressive sensing on the platform of GPU. The reconstruction algorithm based on compressive sensing can achieve a good performance than traditional algorithm, but the process is more complex. With the aid of the GPU acceleration and the redundancy calculation between the adjacent frames, we can achieve real-time video reconstruction result. During the process of acceleration, we divided the whole process into four stages, and find that all the stages are suit for parallel computing. Compared to the sequentialalgorithm, the parallel algorithm achieved a speed up of 35 times. Excepted for GPU acceleration, some other methods to reduce computation of reconstruction for video-frame is proposed. At last, the result of the parallel algorithm is shown and analyzed.