Block-matching motion estimation apparatus under use of linear systolic array architecture
专利汇可以提供Block-matching motion estimation apparatus under use of linear systolic array architecture专利检索，专利查询，专利分析的服务。并且A block-matching motion estimation apparatus in the form of linear systolic array architecture. The apparatus includes a selecting unit for inputting data of a reference block and other data to delay the data of the reference data and selectively outputting the data; a delaying unit having P-1 delay elements where the first delay element delays and outputs an output of the selecting unit and the second to (P-1)-th delay elements delay and output outputs of just front delay element, respectively, the P-1 delay element being connected in serial; and a mean absolute difference operating unit for simultaneously inputting data of a search area to each pipe by connecting in parallel P pipes where operation elements are disposed in a linear systolic array architecture, and then calculating a value of mean absolute difference in each of the search areas by inputting outputs of the selecting unit, the first delay element, and the (P-1)-th delay element to the first pipe, the second pipe, and the (P-th) pipe, respectively.，下面是Block-matching motion estimation apparatus under use of linear systolic array architecture专利的具体信息内容。
What is claimed is:1. A block-matching motion estimation apparatus under use of a linear systolic array architecture, for which a basic search area is (2P.times.N), wherein P is an integer greater than 1 and N is a multiple of 2, said apparatus comprising:means for simultaneously inputting even column search area data and odd column search area data separately into one end of every one of plural linear systolic arrays from a memory;selecting means for selectively inputting reference block data, delaying said reference block data, and inputting the delayed input reference block data into another end of a first one of said linear systolic arrays;delaying means having P-1 delay element for supplying further delayed input reference block data to each remaining one of said linear systolic arrays, wherein a first delay element delays and outputs an output of said selecting unit and second to (P-1)-th delay elements delay and output outputs of an immediately preceding delay element, respectively, said P-1 delay elements being connected serially; andmean absolute difference operating means having P linear systolic arrays, wherein each linear systolic array includes operation elements disposed in a linear systolic array architecture, for calculating means absolute difference values in a 2.times.N search area of the entire 2P by N search area.2. A block matching motion estimation apparatus under use of a linear systolic array architecture as claimed in claim 1, wherein said mean absolute difference operating means comprises:a delay element for delaying input reference block data by two N-1 clock cycles;a selecting device for selectively inputting said input and delayed reference block data; andN operation elements disposed in linear systolic array architecture.
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to a block-matching motion estimation apparatus used with a linear systolic array architecture for a full-search motion estimation operation, or a hierarchical-search motion estimation operation.
2. Description of the Related Art
The motion estimation apparatus according to the prior art is limited in its search range. Therefore, the motion estimation apparatus according to the prior art has had problems in that it is incongruent with any system which must search a wide range. Therefore, in order to perform a motion estimation operation for a wide search range, it is desirable that a plurality of chips are used in parallel by dividing the total motion search areas, or it is desirable that the system is implemented based on a method that one chip repeatedly executes the search operation for each search area. However, under such a situation, there may arise problems in that hardware is very large in its size and time taken in performing such operations is very long.
Further, in the motion estimation apparatus according to the prior art, because the operation speed of internal operation elements is the same as the speed at which data is provided from outside, there may arise therein another problem in that the speed of the internal operation element is limited by the external data supplying speed.
SUMMARY OF THE INVENTION
It is therefore, an object of the invention to provide a block-matching motion estimation apparatus capable of providing a wider search area by using a linear systolic array architecture, as compared with the prior art, reducing the speed of external data supply without slowing down the internal operation of processing elements, and thus improving the overall motion estimation function.
To achieve the above object, the block-matching motion estimation apparatus according to the present invention comprises selecting means for inputting both delayed and undelayed data of a reference block, and then selectively outputting the inputted data in turn; delaying means having P-1 delay elements where the first delay element delays and outputs an output of the selecting means, and the second through {(P-1)-th} delay elements delay and output outputs of the immediately preceding delay elements, respectively, the P-1 delay element being connected in serial; and a mean absolute difference operating means for simultaneously inputting data of a search area to each pipe by connecting in parallel P pipes where operation elements are disposed in a linear systolic array architecture, and then calculating a value of mean absolute difference in each of the search areas by inputting outputs of the selecting means, the first delay element, and the {(P-1)-th} delay element to the first pipe, the second pipe, and the (P-th) pipe, respectively.
BRIEF DESCRIPTION OF THE DRAWINGS
The above and other objects, advantages and features of the present invention will be more apparent from the following detailed description taken with the attached drawings in which:
FIG. 1 is a configuration diagram of a basic operation element used for calculating a value of mean absolute difference during a motion estimation operation according to the present invention;
FIG. 2 is a pipe configuration diagram of a linear systolic array architecture using a basic operation element according to the present invention; and
FIG. 3 is a configuration diagram of a block-matching motion estimation apparatus under use of a linear systolic array architecture according to the present invention.
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT
In the following description, the specific details are set forth to provide a more thorough understanding of the present invention. It will be apparent, however, to one skilled in the art, that the present invention may be practiced without these specific details. In other instances, well known features have not been described in detail so as not to obscure the present invention.
FIG. 1 is a configuration diagram of an operation element for calculating a value of MAD (Mean Absolute Difference) between two adjacent displacement locations in a search area.
Referring to FIG. 1, the operation element has a first shift register 11, second and third shift registers 12 and 13, a MUX 14, a subtracter 15, an absolute value adder 16, and first and second registers 17 and 18. The first shift register 11 stores one of pixel values of a reference block, and also shifts a pixel value of the stored reference block to an adjacent operation element every clock. The second and third shift registers 12 and 13 store pixel values of even and odd columns of the search area, and also shift the pixel values of the stored even and odd columns to the adjacent operation element every clock. The MUX 14 selects either the even column or the odd column of the search area outputted from the second and third shift registers 12 and 13. The subtracter 15 inputs outputs of the first shift register 11 and the MUX 14, and then obtains a difference value from the two pixel values. The absolute value adder 16 receives an output of the subtracter 15, takes an absolute value of the output and adds it to the intermediate value shifted out from the shift register. 18, to thereby obtain the total value. The first and second registers 17 and 18 store values of the MAD in the even and odd columns, respectively.
The operation element performs an operation for one pixel value every clock cycle, and also alternately executes another operation for the MAD in the even and odd columns every clock. Therefore, it takes 128 clock cycles for a calculation operation of the MAD for the reference block consisting of 64 pixels, i.e., 8×8, so that it is possible to obtain two MAD values for two horizontally contiguous displacement locations. In the meantime, as shown in FIG. 2, MAD values for 2N displacement locations can be calculated by connecting N operation elements in a linear manner.
Table <1>shows an index for data of the reference block and the search area. Table <2>shows a data flow and timed operation of each operation element when the N is given as 4.
<TABLE 1>__________________________________________________________________________(a) data of the reference blockC(0,0) C(0,1) C(0,2) C(0,3)C(1,0) C(1,1) C(1,2) C(1,3)C(2,0) C(2,1) C(2,2) C(2,3)C(3,0) C(3,1) C(3,2) C(3,3)(b) data of the search regionS(0,0) S(0,1) S(0,2) S(0,3) S(0,4) S(0,5) S(0,6) S(0,7) S(0,8) S(0,9)S(1,0) S(1,1) S(1,2) S(1,3) S(1,4) S(1,5) S(1,6) S(1,7) S(1,8) S(1,9)S(2,0) S(2,1) S(2,2) S(2,3) S(2,4) S(2,5) S(2,6) S(2,7) S(2,8) S(2,9)S(3,0) S(3,1) S(3,2) S(3,3) S(3,4) S(3,5) S(3,6) S(3,7) S(3,8) S(3,9)S(4,0) S(4,1) S(4,2) S(4,3) S(4,4) S(4,5) S(4,6) S(4,7) S(4,8) S(4,9)S(5,0) S(5,1) S(5,2) S(5,3) S(5,4) S(5,5) S(5,6) S(5,7) S(5,8) S(5,9)S(6,0) S(6,1) S(6,2) S(6,3) S(6,4) S(6,5) S(6,6) S(6,7) S(6,8) S(6,9)S(7,0) S(7,1) S(7,2) S(7,3) S(7,4) S(7,5) S(7,6) S(7,7) S(7,8) S(7,9)__________________________________________________________________________
<TABLE 2>__________________________________________________________________________PE 3 PE 2 PE 1 PE 0path-- e path-- 0 path-- e path-- 0 path-- e path-- 0 path-- e path-- 0__________________________________________________________________________0 S(0,0)1 S(0,0) S(0,0)2 S(1,0) S(0,0) S(0,0)3 S(1,0) S(1,0) S(0,0) S(0,0)4 S(2,0) S(1,0) S(1,0) C(0,0)- S(0,0)5 S(2,0) S(2,0) C(0,0)- S(1,0) S(1,0)6 C(0,0)- S(2,0) C(1,0)- S(2,0) S(1,0)7 C(0,0)- S(3,0) C(1,0)- S(2,0) S(3,0) S(2,0)8 S(4,0) S(0,1) C(1,0)- S(3,0) C(2,0)- S(3,0) S(2,0)9 C(1,0)- S(0,1) S(4,0) S(0,1) C(2,0)- S(3,0) S(4,0) S(3,0)10 S(5,0) S(1,1) C(2,0)- S(0,1) S(4,0) S(0,1) C(3,0)- S(4,0) S(3,0)11 C(2,0)- S(1,0) S(5,0) S(1,1) C(3,0)- S(0,1) S(4,0) C(0,0)- S(5,0) S(4,0) S(0,1)12 S(6,0) S(2,1) C(3,0)- S(1,1) S(5,0) C(0,0)- S(4,0) C(0,10)- S(5,0) S(1,1) S(0,1)13 C(3,0)- S(2,1) S(6,0) C(0,0)- S(5,0) C(0,10)- S(5,0) C(1,0)- S(6,0) S(2,1) S(1,1) S(1,1)14 C(0,0)- S(6,0) C(0,1)- S(6,0) C(1,0)- S(5,0) C(1,1)- S(3,1) S(2,1) S(2,1) S(1,1)15 C(1,0)- S(6,0) C(1,1)- S(6,0) C(2,0)- S(3,1) S(3,1) S(2,1) S(2,1)16 S(0,2) C(1,0)- C(1,1)- C(2,0)- S(6,0) C(2,1)- S(4,1) S(3,1) S(3,1) S(2,1)17 S(0,2) C(1,1)- S(0,2) C(2,0)- C(2,1)- C(3,0)- S(4,1) S(4,1) S(3,1) S(3,1)18 S(1,2) C(2,0)- S(0,2) C(2,1)- S(0,2) C(3,0)- C(3,10)- S(5,1) S(4,1) S(4,1) S(3,1)19 S(1,2) C(2,1)- S(1,2) C(3,0)- S(0,2) C(3,1)- C(0,1)- S(4,1) S(5,1) S(5,1) S(4,1) S(0,2)20 S(2,1) C(3,0)- S(1,2) C(3,1)- C(0,1)- S(5,1) C(0,2)- S(4,1) S(6,1) S(5,1) S(1,2) S(0,2)21 S(2,1) C(3,1)- C(0,1)- S(6,1) C(0,2)- S(5,1) C(1,1)- S(5,1) S(6,1) S(2,2) S(1,2) S(1,2)22 C(0,1)- C(0,2)- S(6,1) C(1,1)- S(6,1) C(1,2)- S(5,1) S(3,2) S(2,2) S(2,2) S(1,2)23 C(0,2)- C(1,1)- C(1,2)- S(6,1) C(2,1)- S(6,1) S(3,2) S(3,2) S(2,2) S(2,2)24 C(1,1)- S(0,3) C(1,2)- C(2,1)- C(2,2)- S(6,1) S(4,2) S(3,2) S(3,2) S(2,2)25 C(1,2)- S(0,3) C(2,1)- S(0,3) C(2,2)- C(3,1)- S(4,2) S(4,2) S(3,2) S(3,2)26 C(2,1)- S(1,3) C(2,2)- S(0,3) C(3,1)- S(0,3) C(3,2)- S(5,2) S(4,2) S(4,2) S(3,2)27 C(2,2)- S(1,3) C(3,1)- S(1,3) C(3,2)- S(0,3) S(4,2) C(0,2)- S(5,2) S(5,2) S(4,2) S(0,3)28 C(3,1)- S(2,3) C(3,2)- S(1,3) S(5,2) C(0,2)- S(4,2) C(0,3)- S(6,2) S(5,2) S(1,3) S(0,3)29 C(3,2)- S(2,3) S(6,2) C(0,2)- S(5,2) C(0,3)- S(5,2) C(1,2)- S(6,2) S(2,3) S(1,3) S(1,3)30 C(0,2)- S(6,2) C(0,3)- S(6,2) C(1,2)- S(5,2) C(1,3)- S(3,3) S(2,3) S(2,3) S(1,3)31 C(0,3)- C(1,2)- S(6,2) C(1,3)- S(6,2) C(2,2)- S(3,3) S(3,3) S(2,3) S(2,3)32 S(0,4) C(1,2)- C(1,3)- C(2,2)- S(6,2) C(2,3)- S(4,3) S(3,3) S(3,3) S(2,3)33 S(0,4) C(1,3)- S(0,4) C(2,2)- C(2,3)- C(3,2)- S(4,3) S(4,3) S(3,3) S(3,3)34 S(1,4) C(2,2)- S(0,4) C(2,3)- S(0,4) C(3,2)- C(3,2)- S(5,3) S(4,3) S(4,3) S(3,3)35 S(1,4) C(2,3)- S(1,4) C(3,2)- S(0,4) C(3,3)- C(0,3)- S(4,3) S(5,3) S(5,3) S(4,3) S(0,4)36 S(2,4) C(3,2)- S(1,4) C(3,3)- C(0,3)- S(5,3) C(0,4)- S(4,3) S(6,3) S(5,3) S(1,4) S(0,4)37 S(2,4) C(3,3)- C(0,3)- S(6,3) C(0,4)- S(5,3) C(1,3)- S(5,3) S(6,3) S(2,4) S(1,4) S(1,4)38 C(0,3)- C(0,4)- S(6,3) C(1,3)- S(6,3) C(1,4)- S(5,3) S(3,4) S(2,4) S(2,4) S(1,4)39 C(0,4)- C(1,34) C(1,4)- S(6,3) C(2,3)- S(6,3) S(3,4) S(3,4) S(2,4) S(2,4)40 C(1,3)- S(0,5) C(1,4)- C(2,3)- C(2,4)- S(6,3) S(4,4) S(3,4) S(3,4) S(2,4)41 C(1,4)- S(0,5) C(2,3)- S(0,5) C(2,4)- C(3,3)- S(4,4) S(4,4) S(3,4) S(3,4)42 C(2,3)- S(1,5) C(2,4)- S(0,5) C(3,3)- S(0,5) C(3,4)- S(5,4) S(4,4) S(4,4) S(3,4)43 C(2,4)- S(1,5) C(3,3)- S(1,5) C(314)- S(0,5) S(4,4) C(0,4)- S(5,4) S(5,4) S(4,4) S(0,5)44 C(3,3)- S(2,5) C(3,4)- S(1,5) S(5,4) C(0,4)- S(4,4) C(0,5)- S(6,4) S(5,4) S(1,5) S(0,5)45 C(3,4)- S(2,5) S(6,4) C(0,4)- S(5,4) C(0,5)- S(5,4) C(1,4)- S(6,4) S(2,5) S(1,5) S(1,5)__________________________________________________________________________
The data of the search area is divided into two. The data of the even column and that of the odd column are provided to the operation elements PE-- 0˜PE-- 3 through paths path-- e and path-- 0, respectively. When a point of time is given as 0, i.e., T=0, a first data (S(0,0)) of the search area is applied to a unit operation element PE-- 3 from outside. Thereafter, one new data is provided for two cycles. On the other hand, the data of the reference block, as shown in FIG. 2, is provided in opposite direction to that of the search area. When a point of time is given as N-1, i.e., T=N-1 (referred to as T=3, in Table <2>), the first data is supplied.
In the same manner as the data of the search area, one new data of the reference block are provided for two cycles. Thereby, the data which is changed in every clock cycle under use of a delay element 21 and a MUX 22, is generated, and is then provided to each operation element. Meantime, after initial clock cycles during which the necessary data is supplied is completed, all operation elements achieve 100 percent utilization efficiency. Further, the successive operations have maximum operation efficiency by being made as pipe line. Notice here again, the supplying speed of the data from outside is half as fast as the operating speed of the internal operation element.
FIG. 3 is a configuration diagram of a block-matching motion estimation apparatus under use of a linear systolic array architecture according to the present invention.
As shown in FIG. 3, the data of the search area is simultaneously provided to the all pipes and that of reference block is provided through delay elements 33 which are constantly disposed between each pipe. At this point, each of the operation elements of a pipe 0 34 is intended to calculate the value of the MAD between two locations adjacent in the horizontal direction, and also is intended to calculate another value of the MAD for N locations in the vertical direction.
For example, the operation element 0 of the pipe 0 having four operation elements is employed to calculate the values of MAD(0, 0) and MAD(0, 1), the operation element 1 the values of MAD(1, 0) and MAD(1, 1), the operation element 2 the values of MAD(2, 0) and MAD(2, 1), and the operation element 3 the values of MAD(3, 0) and MAD(3, 1), respectively.
Also, each of the operation elements of the pipe 1 is subjected to calculate the value of the MAD between two columns which are adjacent to each other in the horizontal direction, i.e, MAD(0˜3, 2) and MAD(0˜3, 3).
Accordingly, in the motion estimation operation architecture consisting of P pipes, as shown in FIG. 3, the value of the MAD for N×2P displacement locations can be efficiently calculated.
In drawings, a reference numeral 32 represents the MUX for selectively outputting one of two input data per clock cycle in turns and reference numerals 31 and 33 represent delay elements having the sizes of 2N and 4N, respectively.
Therefore, according to the block-matching motion estimation apparatus of the present invention, it is possible for the operation element to calculate the values of the MADs within short time. Further, in accordance with the block-matching motion estimation apparatus of the present invention, the supplying speed of the external data is half as fast as the operable speed of the internal operation element. Furthermore, in the present invention, all operation elements achieve 100 percent utilization efficiency after the initial data is provided. Accordingly, there is provided an efficiency in the present invention in that a wider area can be searched, as compared with the prior art. Moreover, according to the present invention, there is provided further efficiency in that the speed of external data supply is reduced without slowing down internal operation of processing elements and the overall motion estimation operation is improved.
Therefore, it should be understood that the present invention is not limited to the particular embodiment disclosed herein as the best mode contemplated for carrying out the present invention, but rather that the present invention is not limited to the specific embodiments described in this specification except as defined in the appended claims.
| 标题 | 发布/更新时间 | 阅读量 |
|---|---|---|
| 帧间运动估计方法、装置及电子设备 | 2020-05-13 | 589 |
| 运动估计的方法、装置、电子设备及计算机可读存储介质 | 2020-05-08 | 712 |
| 石化工厂巡检机器人环境建模与地图构建装置和方法 | 2020-05-08 | 577 |
| 一种运动目标追踪方法、装置、计算机设备及存储介质 | 2020-05-11 | 151 |
| 一种3D-HEVC深度图运动估计并行实现方法及结构 | 2020-05-12 | 311 |
| 基于深度神经网络的视频压缩方法 | 2020-05-15 | 498 |
| 一种融合语义信息的路标定位方法 | 2020-05-08 | 950 |
| 一种基于FPGA的机载吊舱实时图像场景锁定装置及方法 | 2020-05-08 | 462 |
| 一种视频降噪方法和装置 | 2020-05-08 | 576 |
| 一种基于图像中边沿点信息的单目视觉测程方法 | 2020-05-15 | 201 |
高效检索全球专利专利汇是专利免费检索,专利查询,专利分析-国家发明专利查询检索分析平台，是提供专利分析，专利查询，专利检索等数据服务功能的知识产权数据服务商。
我们的产品包含105个国家的1.26亿组数据，免费查、免费专利分析。
专利汇分析报告产品可以对行业情报数据进行梳理分析，涉及维度包括行业专利基本状况分析、地域分析、技术分析、发明人分析、申请人分析、专利权人分析、失效分析、核心专利分析、法律分析、研发重点分析、企业专利处境分析、技术处境分析、专利寿命分析、企业定位分析、引证分析等超过60个分析角度，系统通过AI智能系统对图表进行解读，只需1分钟，一键生成行业专利分析报告。
