VLSI architecture, in particular for motion estimation applications

• US 20020097343A1
• Filed: 09/06/2001
• Published: 07/25/2002
• Est. Priority Date: 09/07/2000
• Status: Active Grant

• Current Assignee: STMicroelectronics SRL (STMicroelectronics NV)
• Original Assignee: STMicroelectronics SRL (STMicroelectronics NV)
• Inventors: Cenciotti, Andrea, Saponara, Sergio, Panucci, Luca, Pau, Danilo, Alfonso, Daniele, Rovati, Fabrizio
• Granted Patent: US 6,724,823 B2
• Time in Patent Office: Days
• US Class Current: 348/700
• CPC Class Codes: G06T 2207/10016 Video; Image sequence
• CPC Class Codes: G06T 7/238 using non-full search, e.g. three-step search

Abstract

A VLSI architecture adapted to be implemented in the form of a reusable IP cell and including a motion estimation engine, configured to process a cost function and identify a motion vector which minimizes the cost function, an internal memory configured to store the sets of initial candidate vectors for the blocks of a reference frame, first and second controllers to manage the motion vectors and manage an external frame memory, a reference synchronizer to align, at the input to the estimation engine, the data relevant to the reference blocks with the data relevant to candidate blocks coming from the second controller, and a control unit for timing the units included in the architecture and the external interfacing of the architecture itself.

Claims

1. A VLSI architecture, particularly for motion estimation applications of video sequences having subsequent frames organized in blocks by means of identification of motion vectors that minimize a given cost function, the architecture adapted to cooperate with an external frame memory and comprising:

a motion estimation engine configured to process the cost function and identify a motion vector that minimizes the cost function;

an internal memory configured to store sets of initial candidate vectors for blocks of a reference frame;

a first controller to manage motion vectors, the first controller configured to provide, starting from the internal memory, the motion estimation engine with the sets of initial candidate vectors and to update the internal memory with the motion vectors identified by the -motion estimation engine;

a second controller to manage the external frame memory, the second controller configured to provide the motion estimation engine with the candidate blocks; and

a reference synchronizer to align, at the input of the motion estimation engine, data relevant to the blocks of the reference frame with data relevant to the candidate blocks coming from the second controller.

2. The architecture of claim 1, further comprising a control unit for timing of the architecture and external interfacing of the architecture.

3. The architecture of claim 1, wherein the estimation engine is configured to store and compare the minimum value of the cost function with at least one given threshold.

4. The architecture of claim 3, wherein the at least one data threshold can be programmed selectively.

5. The architecture of claim 1, wherein the internal memory is of a size parametrizable according to at least one of the parameters selected in the group consisting of:

number of blocks in each frame, and

number of bits used to code the motion vectors.

6. The architecture of claim 1, wherein the motion estimation engine comprises at least one interpolation unit for the implementation of a search of the motion vector with half-pixel accuracy.

7. The architecture of claim 6, wherein the interpolation unit operates on the data deduced by the frame memory.

8. The architecture of claim 6 wherein the interpolation unit is organized in the form of a plurality of subunits.

9. The architecture of claim 1, wherein the motion estimation engine comprises a respective plurality of elements to process the said cost function.

10. The architecture of claim 8, wherein the number of plurality of elements is parametrizable to permit scalability of the architecture.

11. The architecture of claim 9, wherein the respective plurality of elements is followed by an adder tree and by an accumulator for partial sums.

12. The architecture of claim 1, wherein the motion estimation engine comprises a detection unit that detects the minimum value of the cost function and a relevant motion vector.

13. The architecture of claim 1, wherein the motion estimation engine comprises a comparison unit configured to compare a residual matching error with a respective threshold.

14. The architecture of claim 13, wherein the respective threshold can be programmed selectively.

15. The architecture of claim 1, wherein the second controller integrates a memory for temporary storage of the pixels of predictive candidates.

16. A reusable cell architecture, comprising:

a motion estimation engine configured to process a cost function and to identify a motion vector that minimizes the cost function;

an internal memory configured to store sets of initial candidate vectors for blocks of a reference frame;

a first controller to manage motion vectors, the first controller configured to provide, starting from the internal memory, the motion estimation engine with the sets of initial candidate vectors and to update the internal memory with the motion vectors identified by the estimation engine;

a second controller to manage an external frame memory, the second controller configured to provide the estimation engine with the candidate blocks; and

a reference synchronizer to align, at the input of the motion estimation engine, data relevant to the blocks of the reference frame with data relevant to the candidate blocks coming from the second controller; and

a control unit for timing the VLSI architecture and external interfacing of the architecture.

17. A reusable cell architecture adapted to cooperate with an external frame memory for use with video sequences having subsequent frames organized in blocks by means of identification of motion vectors that minimize a given cost function, the architecture comprising:

a motion estimation engine configured to process the cost function and to identify a motion vector that minimizes the cost function, the motion estimation engine comprising at least one interpolation unit for the implementation of a search of the motion vectors with half-pixel accuracy;

an internal memory configured to store sets of initial candidate vectors for blocks of a reference frame;

a first controller to manage motion vectors, the first controller configured to provide, starting from the internal memory, the motion estimation engine with the sets of initial candidate vectors and to update the internal memory with the motion vectors identified by the estimation engine;

a second controller configured to manage the external frame memory, the second controller further configured to provide the motion estimation engine with the candidate blocks;

a reference synchronizer to align, at the input of the motion estimation engine, data relevant to the blocks of the reference frame with data relevant to the candidate blocks coming from the second controller; and

a control unit for timing of the architecture and external interfacing of the architecture.

18. A reusable cell architecture adapted to cooperate with an external frame memory for use with video sequences having subsequent frames organized in blocks by means of identification of motion vectors that minimize a given cost function, the architecture comprising:

a motion estimation engine configured to process the cost function and to identify a motion vector that minimizes the cost function, the motion estimation engine comprising at least one interpolation unit for the implementation of a search of the motion vectors with half-pixel accuracy, and further comprising a respective plurality of circuits configured to process the cost function followed by an adder tree and by an accumulator for calculating partial sums;

an internal memory configured to store sets of initial candidate vectors for blocks of a reference frame;

a first controller to manage motion vectors, the first controller configured to provide, starting from the internal memory, the motion estimation engine with the sets of initial candidate vectors and to update the internal memory with the motion vectors identified by the estimation engine;

a second controller configured to manage the external frame memory, the second controller further configured to provide the motion estimation engine with the candidate blocks;

a reference synchronizer to align, at the input of the motion estimation engine, data relevant to the blocks of the reference frame with data relevant to the candidate blocks coming from the second controller; and

a control unit for timing of the architecture and external interfacing of the architecture.

19. A reusable cell architecture for use with video sequences having subsequent frames organized in blocks by means of identification of motion vectors that minimize a given cost function, the architecture adapted to cooperate with an external frame memory, the architecture comprising:

a motion estimation engine configured to process the cost function and to identify a motion vector that minimizes the cost function, the motion estimation engine comprising a detection unit that detects the minimum value of the cost function and a relevant motion vector;

an internal memory coupled to the motion estimation engine and configured to store sets of initial candidate vectors for blocks of a reference frame;

a first controller coupled to the motion estimation engine for managing motion vectors, the first controller configured to provide, starting from the internal memory, the motion estimation engine with the sets of initial candidate vectors and to update the internal memory with the motion vectors identified by the motion estimation engine;

a second controller coupled to the first controller and to the motion estimation engine, the second controller configured to manage the external frame memory and to provide the motion estimation engine with the candidate blocks; and

a reference synchronizer coupled to the motion estimation engine and configured to align at the input of the motion estimation engine data relevant to the blocks of the reference frame with data relevant to the candidate blocks sent from the second controller.

20. A reusable cell architecture for use with video sequences having subsequent frames organized in blocks by means of identification of motion vectors that minimize a given cost function, the architecture adapted to cooperate with an external frame memory, the architecture comprising:

a motion estimation engine configured to process the cost function and to identify a motion vector that minimizes the cost function, the motion estimation engine comprising a detection unit that detects the minimum value of the cost function and a relevant motion vector, and further comprising a comparison unit configured to compare a residual matching error with a respective threshold;

an internal memory coupled to the motion estimation engine and configured to store sets of initial candidate vectors for blocks of a reference frame;

a first controller coupled to the motion estimation engine for managing motion vectors, the first controller configured to provide, starting from the internal memory, the motion estimation engine with the sets of initial candidate vectors and to update the internal memory with the motion vectors identified by the motion estimation engine;

a second controller coupled to the first controller and to the motion estimation engine, the second controller configured to manage the external frame memory and to provide the motion estimation engine with the candidate blocks; and

a reference synchronizer coupled to the motion estimation engine and configured to align at the input of the motion estimation engine data relevant to the blocks of the reference frame with data relevant to the candidate blocks sent from the second controller.