Motion Estimation Hardware

Hardware implementation of Motion Estimation algorithm written in Verilog-HDL.

Table of Contents

1. About Motion Estimation
2. Description
3. Required Tools
4. How to Run Simulation
5. How to Implement Motion Estimation Hardware
6. Directory Structure
7. Behavioral Simulation Result
8. Python Result
9. Block Diagrams
10. Source Files
11. To Do

---

About Motion Estimation

Motion Estimation is a process of finding motion vectors from a reference frame to the current frame. This process involves the following steps for each macro block in the current frame:

• Find a block with the lowest matching error in the search window of the reference frame.
• Use the sum of absolute difference (SAD) between corresponding pixels.
• The displacement between the current macro block and the best matching macro block is the motion vector.

---

Description

• Search Memory Window: 31x31 pixels
• Reference Memory Block: 16x16 pixels
• Parallel Processing: 16 Processing Elements
• Clock Frequency: 31.25 MHz
• Post-Implementation: Nexys-A7-100T FPGA board

---

Required Tools

• Vivado ML Edition 2023.2
• Python 3.10

---

How to Run Simulation

1. Go to the /testbench/testcase/ directory and run the /testbench/testcase/randblock_gen.py script.
2. After running the script, execute /testbench/testcase/min_SAD.py to generate the /testbench/testcase/min_SAD.txt file.
3. Add the /testbench/testcase/SearchWindowMemory_hw.txt, /testbench/testcase/ReferenceBlock_hw.txt, and /testbench/testcase/min_SAD.txt files as simulation files in your VIVADO project.
4. Finally, run the /testbench/ME_tb.v file.

---

How to Implement Motion Estimation Hardware

1. Go to the /constraint directory.
2. Add the constraint file to the relevant VIVADO project as a constraint file.
3. Add the source files from the /rtl directory and set MotionEstimationTop.v as the top module.
4. Run the implementation.

---

Directory Structure

The Motion Estimation Processor source code is organized as follows:

/docs       project report
/constraint constraint file
/rtl        RTL files for Motion Estimation Processor
/testbench  testbench source and testcase files

---

Behavioral Simulation Result

---

Post-Imp Functional Simulation Result

---

Python Result

---

Block Diagrams

Motion Estimation Block Diagram

Motion Estimation RTL Block Diagram

Processing Elements Scheduling

---

Source Files

Verilog Source Files

• SearchWindowMemory.v: Implements the memory that stores the search window data used in the motion estimation process.
• ReferenceBlockMemory.v: Stores the reference block data that is compared against the search window data during motion estimation.
• ControlUnit.v: Manages the control signals for coordinating the operations of different components in the motion estimation hardware.
• Datapath.v: Defines the data path for processing the input data, performing the necessary computations for motion estimation.
• InstantMinComp.v: Computes the minimum value instantaneously, used for determining the best match in motion estimation.
• MinTrackerComp.v: Tracks the minimum values over multiple comparisons to find the optimal motion vector.
• MotionEstimationTop.v: Top-level module that integrates all submodules and coordinates the overall motion estimation process.
• ProcessingElements.v: Contains the array of processing elements that perform the core computations for motion estimation, such as calculating the Sum of Absolute Differences (SAD).

Verilog Testbench Files

• randblock_gen.py: Generates random blocks for use in motion estimation testing.
• min_SAD.py: Computes the minimum Sum of Absolute Differences (SAD) in software for comparison with hardware results.
• ME_tb.v: Testbench for the Motion Estimation hardware. It initializes signals, loads data from text files into memories, and compares hardware and software results.
• SearchWindowMemory_hw.txt: Contains binary data for initializing the search window memory in the testbench.
• ReferenceBlock_hw.txt: Contains binary data for initializing the reference block memory in the testbench.
• SearchWindowMemory_sw.txt: Contains decimal data for initializing the search window memory in the software.
• ReferenceBlock_sw.txt: Contains decimal data for initializing the reference block memory in the software.
• min_SAD.txt: Contains the expected minimum SAD value calculated by the software for comparison with the hardware result.

---

To Do

• Make scalable (parametric) structure: Refactor the design to be scalable and parameterizable to support various configurations.
• Demo on FPGA: Demonstrate the working of the motion estimation hardware on an FPGA.
• Make configurable: Ensure the hardware is configurable to support different parameters and use cases.