MOTION ESTIMATION IMPLEMENTATION IN RECONFIGURABLE PLATFORMS com.tk@fit.ac.cy
OUTLINE MOTION ESTIMATION BLOCK MATCHING MOTION ESTIMATION EXISTING MOTION ESTIMATION IMPLEMENTATIONS 10/12/2010
MOTION ESTIMATION (ME) Video compression kernel based on high correlation between frames in a video sequence Operation that estimates the motion vectors of moving objects in sequence Most computationally complex and power intensive operation in video compression Objects can have complex shapes and motion (rotation, camera zoom etc.) Block matching ME is an efficient approximation 10/12/2010
BLOCK MATCHING ME Συντεταγμένες διανύσματος κίνησης: (x + u), (y + v) 10/12/2010
BLOCK MATCHING ADVANTAGES DISADVANTAGES simple straightforward very efficient DISADVANTAGES Causes block artifacts Can still be computationally intensive for large frame sizes Efficient search strategy is required 10/12/2010
SEARCH ALGORITHMS: FULL SEARCH Exhaustive (raster scan) Optimum result Most computationally intensive 10/12/2010
SEARCH ALGORITHMS: SPIRAL SEARCH Spiral scan with early termination Worst case equal to FS Can be trapped in a local minimum 10/12/2010
SEARCH ALGORITHMS: THREE STEP SEARCH A search algorithm based on a coarse original search using 9 points The search is refined in two subsequent searches around the best match found in each step Total number of search points: 9 + 2x18 = 25 Can be trapped in a local minimum 10/12/2010
MATCHING CRITERION Second most important implementation factor after the search strategy Must be as computationally simple as possible yet provide accurate evaluation of candidate block 10/12/2010
POSSIBLE SEARCH CRITERIA Mean Absolute Error Mean Square Error Sum of Square Differences Sum of Absolute Differences 10/12/2010