wbrueckner / cv2PYNQ-The-project-behind-the-library

This project describes how the cv2PYNQ python library was built

cv2PYNQ - The project behind the library

Board used: Pynq-Z1

Vivado Version: 2017.4

Brief description:

This project accelerates OpenCV-functions on the PYNQ platform. The library implements a specific set of popular image filters and feature detection algorithms. The calculation of time-consuming tasks is implemented in the Programmable Logic (PL) of the ZYNQ chip. All data are transferred through Direct Memory Access and streamed back to the DRAM and thus, the Python program. cv2PYNQ also includes the Video-Subsystem of the base project of PYNQ. Therefore, the HDMI In and Out interfaces can be used in your application. The library calculates every filter for gray-channel images with 1080p within 16 ms if the input and output buffers are located in the contiguous memory of the chip.

Currently accelerated functions:

• Sobel: 3x3; 5x5
• Scharr
• Laplacian: ksize = 1; 3; 5
• blur: ksize = 3
• GaussinBlur: ksize = 3
• erode: ksize = 3
• dilate: ksize = 3
• Canny

Link to YouTube Video: https://www.youtube.com/watch?v=nRxe-NqvOl8

Get Started

Install by typing:

git clone https://github.com/wbrueckner/cv2pynq.git   
cd cv2pynq/   
pip3.6 install -e .   

into the terminal on your Pynq-Z1 board. The library comes with a jupyter notebook to demonstrate its usage and capabilities. You find the notebook in the cv2PYNQ folder of your home tree after installation.

Directory Structure

├── cv2pynq/                        *The python library as git submodule*    
├── cv2PYNQ_vivado/                 *The vivado project*
│   ├── cv2PYNQ.tcl
│   └── cv2PYNQ_vivado.xpr
├── doc/                            *Further documentaion*
│   └── PYNQ-OpenCV_xohw18-155.pdf
├── ip/                             *Used IP in this porject* 
│   ├── HLS/                        *HLS projects for image porcessing cores* 
│   │   ├── canny/
│   │   ├── dilate/
│   │   ├── erode/
│   │   ├── filter2D/
│   │   ├── filter2D_5/
│   │   └── filter2D_f/
│   └── PYNQ/                       *IP cores from the [base](https://github.com/Xilinx/PYNQ) project 
│       ├── ip/         
│       └── pynq_video_subsystem/
└── hw/                             *Bitstream and TCL-File*

Contribute to cv2PYNQ

The library was build using Vivado 2017.4 and Vivado HLS 2017.4. It was designed with the need for expansion in mind, so we created a very adaptable architecture. You can enhance the capabilities by adding image processing cores in the image_filters subsystem of the Vivado project. Every image processing core is created with Vivado HLS.

• Clone the project and open the "cv2PYNQ_vivado.xpr" project file.
• Just take the filter2D project as a draft and exchange the hls::Filter2D-function with the new functionality.
• Add the new IP to the IP Catalog of the Vivado project.
• Insert it into the image_filters sub-block and connect the IP like the others.
• Expand the number of interfaces in the AXI Interconnect and the two AXI-Stream Interconnects.
• Assign an address in the Address Editor.
• Run the Implementation. Watch out for a Total Negative Slack (TNS) of >= 0ns.
• Adapt the __init__.py and cv2pynq.py files with the added function in the library.
• Test the newly added functionality.
• Contribute to this project.