High-speed motion estimation method

Abstract

A high-speed motion estimation method for estimating a motion of a picture at high speed in a high-speed motion estimation apparatus including an integer-times motion estimation unit and a decimal-times motion estimation unit. The integer-times motion estimation unit (1) receives pixel data of a picture for estimation of motion from an external frame memory, (2) stores pixel data of a current region in a current macroblock storage where multiple inputs/outputs are made, (3) stores pixel data of an estimation region in an estimation region macroblock storage of a sharable multiple input/output bank type, and (4) predicts an integer-times motion vector using both stored pixel data. The decimal-times motion estimation unit reads the pixel data of the estimation region from the estimation region macroblock storage which is sharable, and predicts a decimal-times motion vector using the read pixel data of the estimation region and the integer-times motion vector.

What is claimed is: 
     
         1 . A high-speed motion estimation method for estimating a motion of a picture at high speed in a high-speed motion estimation apparatus including an integer-times motion estimation unit and a decimal-times motion estimation unit, the high-speed motion estimation method comprising:
 receiving, by the integer-times motion estimation unit, pixel data of a picture for estimation of a motion from an external frame memory;   storing, by the integer-times motion estimation unit, pixel data of a current region in a current macroblock storage where multiple inputs/outputs are made, and storing pixel data of an estimation region in an estimation region macroblock storage of a sharable multiple input/output bank type, among the pixel data of the picture;   predicting, by the integer-times motion estimation unit, an integer-times motion vector by using the stored pixel data of the current region and the stored pixel data of the estimation region;   reading, the decimal-times motion estimation unit, the pixel data of the estimation region from the estimation region macroblock storage which is sharable; and   predicting, by the decimal-times motion estimation unit, a decimal-times motion vector by using the read pixel data of the estimation region and the integer-times motion vector.   
     
     
         2 . The high-speed motion estimation method of  claim 1 , wherein the predicting of an integer-times motion vector comprises:
 reading the stored pixel data of the current region and the stored pixel data of the estimation region by skipping one section;   performing 2:1 subsampling on the read pixel data of the current region and the read pixel data of the estimation region, respectively;   calculating a Sum of Absolute Differences (SAD) with the 2:1 subsampled current macroblock and the 2:1 subsampled macroblock of the estimation region; and   predicting the integer-times motion vector by using the minimum value of the calculated SAD values.   
     
     
         3 . The high-speed motion estimation method of  claim 2 , wherein the calculating of a SAD comprises:
 calculating a sum of upper absolute differences for predicting motion estimation of an upper portion which is above a predetermined reference point to calculate a first minimum SAD value; and   calculating a sum of lower absolute differences for predicting motion estimation of a lower portion which is above a predetermined reference point to calculate a second minimum SAD value at same time with the calculation of the sum of the upper absolute differences   
     
     
         4 . The high-speed motion estimation method of  claim 1 , wherein the reading of the pixel data of the estimation region comprises reading the pixel data of the estimation region when an estimation region sharable signal is received from the integer-times motion estimation unit. 
     
     
         5 . The high-speed motion estimation method of  claim 1 , wherein the predicting of a decimal-times motion vector comprises:
 calculating Sum of Absolute Differences (SAD) with the read pixel data of the estimation region and the integer-times motion vector; and   predicting the decimal-times motion vector by using the minimum value of the calculated SAD values.