REAL-TIME MOTION ESTIMATION FOR 1080P VIDEOS ON GRAPHICS PROCESSING UNITS WITH SHARED MEMORY OPTIMIZATION

摘要

Motion Estimation plays an important role in many applications. A shared-memory-optimized implementation of motion estimation is studied in this work, where bank conflict is minimized and bank occupancy is maximized. Experimental results shows that at most 38 times speedup can be achieved compared with a 3GHz CPU. Moreover, based on the optimization technique, a fast algorithm for real-time motion estimation on 1080p videos is also proposed in this paper.