Computing 2D Homography Matrix from a pair of images is essential for many computer vision application like image/video stitching, high resolution images. Computing Homography Matrix involves compute intensive algorithms like SURF/SIFT, RANSAC. Recently a CNN based approach is suggested which directly estimates the Homography matrix from a pair of images once the network is trained. CGRAs are power efficient parallel architectures suitable for embedded systems due to their low power budget.In this presentation we briefly describe DRRA, a CGRA architecture and describe mapping Homography CNN layers on it. DRRA resource requirements is estimated to achieve the performance.