权利要求

BACKGROUND OF THE INVENTION

The present invention relates to an apparatus for estimating motion and a method thereof, and more particularly to an apparatus for estimating real-time motion in a motion-compensated predictive coder for data compression of a motion image and a method suitable for the apparatus.

A predictive coding technique is utilized to heighten data compression efficiency during transmitting or recording of motion image data on a recording medium. Especially in data having great motion, a motion-compensated predictive coding technique is generally employed to prevent the degradation of the data compression efficiency which is caused by a simple predictive coding. In the motion-compensated predictive coding, an error signal is not solely transmitted or recorded as in the simple predictive coding, but motion information is extracted to thereby transmit and record both the error signal and motion information. Here, the block most similar to a certain block of a current frame is searched in a fixed search area of a previous frame, so that the searched block is used as a predicted value during the predictive coding. The search area is constituted by all the surrounding areas centering around the certain block of the current frame, and the relative distance from the current position to the position of the similar block within the search area is termed as a "motion vector" or "displacement vector." That is, the motion vector is the relative distance of the current block position and the position of the most similar previous block. Thus, an absolute value of the motion vector becomes large when the motion is great, it becomes small when the motion is slight, and becomes (0, 0) when there is no motion. The search of the motion vector of the previous block having a minimum error value is called as a motion estimation. and generally adopts a block-matching algorithm (BMA). The BMA commonly utilizes a full search technique wherein the most similar previous block is obtained by a comparative operation in the ratio of 1:1 with the current block, with respect to all blocks formable in the search area. Such a BMA is disclosed in "IEEE Transactions on Circuits and Systems for Video Technology", vol.2, nº 2, June 1992, p. 169-175; HSIEH et al.: "VLSI Architecture for Block-Matching Motion Estimation Algorithm". However, since all blocks are compared one-to-one in the full search, both the overflow of the operation and the need for many I/O pins impede the hardware realization of a real-time motion vector estimator.

The similarity calculation of the compared blocks uses a minimum absolute error D(i, j), which can be written as: where a(m, n) designates a current pixel value, b(m+j, n+j) is a previous pixel value, and (i, j) is a motion vector value.

Among the thus-obtained D(i, j) values, the relative position of the previous block having the minimum error value with respect to the current block becomes the motion vector. An example of the operation amount and processing speed required for realizing the hardware for real-time motion estimation can be give as a set of circumstances:

• data rate: 13.5MHz
• current block size: 256 pixels (16×16)
• motion vector range: -8≦i, j≦7
• search window size: 961 pixels (31×31)
• previous block size: 256 pixels (16×16)
• number of the previous block: 256

Here, the total operating number of the full search technique is 256×256 for each block, and 256 operations are required for each pixel. Since a subtraction operation, absolute value calculation and accumulation must be performed for each operation, 256 operations are necessary for each pixel period, so that the processing requires approximately 0.29ns per data. Therefore, it can be appreciated that, when the processing is carried out per data, real-time processing is difficult due to the relation between the required total number of operations and the processing speed.

SUMMARY OF THE INVENTION

The present invention is contrived to solve the above-described problem of the prior art. It is an object of the present invention to provide an apparatus for estimating real-time motion of a motion-compensated predictive coder capable of real-time processing, and a method suitable for the apparatus.

It is another object of the present invention to provide an apparatus for estimating real-time motion capable of minimizing the number of I/O pins in a single chip.

To achieve the above objects according to the present invention, there is provide a real-time motion estimation apparatus for performing one-to-one comparison of a plurality of previous pixel blocks having a size of Mlines×Ndots within a search window having a size of (2M - 1)lines × (2N - 1)dots with a current pixel block having a size of Mlines×Ndots in order to extract relative position information between the current pixel block and the previous pixel block most similar to the current pixel block as motion information, which comprises:

• previous block formation means for simultaneously inputting first and second serial pixel data strings within the search window, and sequentially outputting in parallel the serial pixel data strings corresponding to the respective previous pixel blocks;
• block-matching algorithm processing means for sequentially inputting the serial pixel data strings and a third senal pixel data string of the current pixel block to simultaneously perform one-to-one comparison of the plurality of previous pixel blocks with the current pixel block in parallel, thereby outputting in parallel the respective absolute error data strings of the plurality of previous blocks with respect to the current pixel block;
• minimum absolute error detecting means for obtaining minimum absolute error data of each group by inputting and dividing the parallel absolute error data stnngs into
• a plurality of groups, and comparing the minimum absolute error data of respective groups with one another, thereby producing minimum absolute error data within the search window;
• motion vector generating means for generating a motion vector corresponding to the minimum absolute error data produced in the search window by being synchronized to the operation of the minimum absolute error detecting means for obtaining the minimum value; and
• output controlling means for outputting the absolute error data of a motion vector (0, 0), the produced minimum absolute error data and the generated motion vector to an I/O bus, and selectively outputting the corresponding absolute error data among a plurality of absolute error data strings in accordance with an external motion vector input via the I/O bus.

That is, through the above-described construction in the present invention, the pixel values of the current block and the previous block within the search are sequentially input to thereby minimize the I/O pin numbers, so that the realization of a VLSI is facilitated. Also, by including processing elements which are the same as the previous block numbers, the operation simultaneously begins and ends in parallel, thereby performing a real time operation. Here, the transmission of pixel data required for each processing element is carried out such that pixel data of the current block is simultaneously transmitted, and the pixel data of the previous block is transmitted by means of a horizontal and vertical pipeline system. By this transmission line structure, the apparatus for estimating motion has a systematic bus structure and the I/O pin number can be minimized. Furthermore, in order to produce a minimum absolute error, the error data is divided into a plurality of groups, the divided error data is simultaneously compared per group, and then each minimum absolute error of each group is compared, thereby producing the final minimum absolute error. Thus, the circuitry is simplified and the signal delay is shortened, which realizes effective hardware.

Moreover, according to the present invention, an error value with respect to a certain motion vector from external unit can be output.

BRIEF DESCRIPTION OF THE DRAWINGS

The above objects and other advantages of the present invention will become more apparent by describing in detail a preferred embodiment thereof with reference to the attached drawings in which:

• FIG. 1 is a view for explaining the concept of motion estimation;
• FIG. 2 is a block diagram showing an apparatus for estimating real time motion according to the present invention;
• FIG. 3 is a circuit diagram of the previous block formation unit shown in FIG. 2;
• FIG. 4 is a circuit diagram showing the construction of the first shift register shown in FIG. 3;
• FIG. 5 is a circuit diagram showing the construction of the second shift register shown in FIG. 3;
• FIG. 6 is a circuit diagram of the block-matching algorithm operation processor shown in FIG. 2;
• FIG. 7 is a circuit diagram showing the construction of a processing element shown in FIG. 6;
• FIG. 8 is a circuit diagram of the minimum absolute error detector shown in FIG. 2;
• FIG. 9 is a circuit diagram of the motion vector generator shown in FIG. 2; and
• FIG. 10 is a circuit diagram of the output controller shown in FIG. 2.

DETAILED DESCRIPTION OF THE INVENTION

In FIG. 1, a reference symbol CB designates a current block size of Mlines×Ndots in the current frame, wherein M=N=16. SW is a search area in a previous frame, i.e., a search window size of (2M-1)lines×(2N-1)dots. Therefore, the search window includes M×N previous blocks (PB) having the same size as current block CB. Search window SW sets the left uppermost pixel position of current block CB as reference coordinates (0, 0), and has respective coordinates from -8 to M+6 in the vertical direction and from -8 to N+6 in the horizontal direction. Accordingly, i and j of motion vector MV(i,j) which represents the relative displacement from the left uppermost pixel position of current block CB to the left uppermost pixel position of each previous block PB, exist within the range of -8≤i, j≤7. The motion estimation is carried out in such a manner that a subtraction operation is performed for the 256 pixels in current block CB and the corresponding 256 pixels in the previous block, and then absolute values of the difference values are obtained, whereby a previous block whose sum of 256 absolute values is minimum is assumed to be in the 145° oblique-lined portion, and motion vector MV(i,j) and a minimum absolute error D(i,j) which is the above sum are provided as motion estimation data.

FIG. 2 illustrates a block diagram showing an apparatus for estimating real-time motion according to the present invention. The apparatus includes a previous block formation unit 100, a block-matching algorithm (BMA) processor 200, a minimum absolute error (MAE) detector 300, a motion vector (MV) generator 400, an output controller 500, and a control signal generator 600.

Previous block formation unit 100 receives first and second serial pixel data strings SPD1 and SPD2 within a search window, and sequentially outputs block pixel data strings PBD in parallel respectively corresponding to all previous blocks PB(i,j) within the search window. Referring to FIG. 3, previous block formation unit 100 includes a first demultiplexer 110, a first latch 120, an M×N first shift register matrix 130, a second demultiplexer 140, a second latch 150, and an (M-1)×N second shift register matrix 160. First demultiplexer 110 sequentially receives the pixel data string of upper sixteen rows of the search window for each column and demultiplexes the received pixel data strings to be supplied to first latch 120. First latch 120 parallel outputs the latched pixel data of the upper sixteen rows by a first predetermined clock period. M×N first shift register matrix 130 leftward-horizontally shifts the parallel pixel data of the upper sixteen rows supplied from first latch 120 by a second predetermined clock period, and upward-vertically shifts them by a third predetermined clock period. Corresponding to each previous block PB within the search window, each first shift register SRA forms the M×N pixel data of the corresponding previous block PB into serial data PBD(x, y) to thereby output the serial data.

Accordingly, M×N first shift register matrix 130 simultaneously outputs each serial data string of M×N previous blocks in parallel. Second multiplexer 140 sequentially inputs pixel data strings of the lower fifteen rows of the search window per column, and demultiplexes the input pixel data strings to be supplied to second latch 150. Second latch 150 parallel outputs latched pixel data of the lower fifteen rows by the first predetermined clock period. (M-1)×N second shift register matrix 160 leftward-horizontally shifts the parallel pixel data of the lower fifteen rows supplied from second latch 150 by the second predetermined clock period, and upward-vertically and recursively shifts them by the third predetermined clock period. The second shift registers SRB at the uppermost line are connected to supply data to the first shift registers SRA at the lowest line of M×N first shift register matrix 160.

Referring to FIG. 4, first shift register SRA has a first flip-flop FF1 for horizontal-shifting, a second flip-flop FF2 for vertical-shifting and a multiplexer MUX1 for selectively outputting the outputs of first and second flip-flops FF1 and FF2.

Referring to FIG. 5, second shift register SRB has a multiplexer MUX2 for selecting data supplied from registers on the right and lower parts of multiplexer MUX2 and a flip-flop FF3 for supplying the data which is selected in multiplexer MUX2 to the registers on the left and upper parts of multiplexer MUX2 according to a clock signal.

Referring to FIG. 6, BMA operation processor 200 includes an M×N processing element PE matrix corresponding to M×N previous blocks. Referring to FIG. 7, each processing element PE has a subtractor SUB for performing the subtraction of a pixel data string CBD of current block CB and pixel data string PBD(x, y) of a previous block PB(x, y), an absolute value producer ABS for obtaining the absolute value of the output from subtractor SUB, an accumulator ACC for accumulating absolute values with respect to M×N pixel data. and an output latch LT1 for outputting the accumulated value of accumulator ACC as an error BE(i, j) with respect to current block CB off previous block PB. Accumulator ACC has an adder ADD and a vertical latch LT2.

Therefore, BMA operation processor 200 simultaneously generates M × N errors BE(i, j) of M × N previous blocks PB with respect to current blocks CB by real-time processing.

Referring to FIG. 8. MAE detector 300 has four group minimum detectors 310, 320, 330 and 340 to produce group minimum error per group which are obtained by dividing the error into four groups, a first group of BE(1, 1) ∼ BE($\frac{\text{M}}{\text{2}}$, $\frac{\text{N}}{\text{2}}$, a second group of BE(1, $\frac{\text{N}}{\text{2}}$ + 1) ∼ BE($\frac{\text{M}}{\text{2}}$ + 1, N), a a third group of BE($\frac{\text{M}}{\text{2}}$ +1, 1) ∼ BE(M, $\frac{\text{N}}{\text{2}}$ + 1) and a fourth group of BE($\frac{\text{M}}{\text{2}}$ +1, $\frac{\text{N}}{\text{2}}$+1) ∼ BE(M, N). Each group minimum detector includes a first multiplexer MUX3 for parallel inputting error BE of the corresponding group to thereby generate serial error data strings ERA, ERB, ERC and ERD in response to an error selection signal BES, a second multiplexer MUX4 for selecting the serial error data string and group minimum error data supplied from another group in response to a group selection signal GPS, and a comparator COMP for initially inputting to compare a control signal ICP for selecting the first error value of each group with the output of second multiplexer MUX4, and latching the smaller value between two inputs according to the result of the comparison. First group minimum detector 310 further includes a third multiplexer MUX5 for selecting the minimum value of the second group detected by means of second group minimum detector 320 and the minimum value of the fourth group detected by mens of fourth group minimum detector 340, to thereby supply the selected value to second multiplexer MUX4 thereof as the minimum value of another group. The minimum value is obtained by means of the comparison between the first and second groups, between the second and fourth groups, between the third and first groups and between the fourth and the third groups. Then, the final outputs of the first and fourth groups are compared with each other, so that a final minimum absolute error MAE is output via an output latch LT3. Output latch LT3 responds to an output enable signal OE. Each comparator COMPA through COMPD respectively output selection signals GSA through GSD corresponding to the minimum value determined after the comparison.

Therefore, in order to speed up the comparing detection, MAE detector 300 performs the comparison per four groups, and outputs final minimum absolute error MAE by comparing the minimum values of respective blocks with each other per blocks.

Referring to FIG. 9, MV generator 400 includes a motion vector (MV) look-up table 410 for sequentially generating the motion vectors corresponding to previous blocks PB of each group, group motion vector selectors 420, 430, 440 and 450 respectively corresponding to the group minimum detectors of MAE detector 300, and an output latch LT5. The group MV selectors respectively include a first multiplexer MUX6 for selecting the motion vectors supplied from MV look-up table 410 and another group in response to group selection signal GPS, a second multiplexer MUX7 for selecting the input motion vector and the latched motion vector in response to each selection signal GSA through GSD supplied from MAE detector 300, and a latch LT4 for latching the motion vector selected in second multiplexer MUX7. First group MV selector 420 further has a third multiplexer MUX8 for selecting one between motion vectors selected from the second group and another group to thereby supply the selected motion vector to first multiplexer MUX6 thereof as the motion vector of another group. The output from second multiplexer MUX7 of first group MV selector 420 is supplied to output latch LT5. The output from latch LT4 of first group MV selector 420 is supplied to third group MV selector 440, and the output from latch LT4 of third group MV selector 440 is supplied to fourth group MV selector 450. The output from latch LT4 of third group MV selector 440 is supplied to fourth group MV selector 450, and the output from latch LT4 of fourth group MV selector 440 is supplied to second and first group MV selectors 430 and 420.

Thus, MV generator 400 selects the previous block's motion vector corresponding to the minimum value detected by being synchronized to the minimum value detecting operation of MAE detector 300, thereby simultaneously producing and outputting motion vector MV(i,j) of the previous block PB having the final minimum error and the minimum absolute error.

Referring to FIG. 10, output controller 500 includes an I/O bus interface 510, a counter CNT, a first multiplexer MUX9 and a second multiplexer MUX10. Counter CNT which is a 64-mode counter for counting to 64 by inputting a clock signal, generates error selection signal BES. In response to a group selection signal GPS, first multiplexer MUX9 selects error selection signal BES generated from 64-mode counter CNT and an error selection signal input from an external portion, thereby outputting error selection control signal BES of MAE detector 300. Second multiplexer MUX10 selects block errors ERA, ERB, ERC and ERD of each group from MAE detector 300 in response to the externally input selection signal, and provides the selected one to I/O bus interface 510. I/O bus interface 510 receives the reference block error BE($\frac{\text{M}}{\text{2}}$ +1, $\frac{\text{N}}{\text{2}}$ +1) in case of no motion, minimum absolute error MAE and MV. outputs them to the I/O bus. and provides a certain external motion vector EMV to first and second multiplexers MUX9 and MUX10 as 6-bit and 2-bit signals, respectively. Therefore, output controller 500 not only outputs the MAE and corresponding MV as motion estimating data but also generates a corresponding error by means of an external control signal OE in response to motion vector EMV input from an external unit, i.e., an external selector.

Control signal generator 600 receives a clock signal to thereby supply various control signals to respective units described above.

According to the present invention as described above, in order to minimize the number of I/O pins, a serial input system is utilized for data input from an external unit, all previous blocks and current blocks are simultaneously processed in parallel by means of a pipeline system in internal processing, and a minimum value or each group is detected together with the selection of the corresponding motion vector. As a result, a high processing speed can be achieved which enables the estimation of real-time motion.