Accelerating Phase Based Motion Estimation with Hierarchical Search Technique Using Parallel Threading in Graphical Processing Unit (GPU)

Summary This paper presents Phase Only Correlation (POC) methods in hierarchical search motion estimation for high resolution digital video using Graphical Processing Unit (GPU). Using the POC function, one can estimate the translational displacement as well as the degree of similarity between two image blocks from the location and height of the correlation peak, respectively[1]. Motion Estimation is a process for defining object movement in digital video sequences. Motion Estimation is a system used in some field such as image processing, image analysis, video coding, and computer vision. A POC based hierarchical search is a high cost algorithm results in long processing time, thus the system developed in this paper proceed POC function in Graphical Processing Unit using parallel threading technology. The evaluation counts processing time speed of the methods using Graphical Processing Unit in high definition video with 1280 x 720 pixel resolution. The results show that the methods using GPU performs accelerating speed more than two times faster processing 2 layer hierarchical search in 256x256 POC block size than doing the same methods using CPU. Using the NVidia GeForce 9600GT GPU, kernel execution with 256 thread per block, 9 32-bit register per thread, and 36 bytes of memory shared for every thread block, the multiprocessor maximum occupancy is 100%, with 768 active threads per multiprocessor, 24 Active Warps per multiprocessor, and 3 active thread blocks per multiprosessor.