Utilizing PYNQ for Accelerating Image Processing Functions in ADAS Applications

As reconfigurable platforms become more and more complex through heterogeneous architectures, exploiting the gained computational speedup efficiently remains a difficult task. Such FPGA based platforms like the Xilinx Zynq require detailed knowledge in FPGA design as well as embedded programming and platform configuration which makes it hard to exploit it for embedded applications where usually no reconfigurable hardware was used, e.g. optical ADAS applications. Through the development of the PYNQ system handling the entire platform by a Python interface becomes possible. The framework comes with a wide support of OpenCV functions for ARM-based processing. However, the main PYNQ project does not offer hardware support for these functions. In this paper, we describe the design an easy to use hardware library for applying complex image processing applications on PYNQ. Our library has been published as open-source repository. It supports various common image functions including Gaussian blur, Sobel or Median filter and even the Canny edge detection. Using the library does not require any hardware knowledge. Supported hardware functions are automatically detected and outsourced on the FPGA.With the provided HLS interface our library can be extended by adding further components. Our results show that the library can be used for accelerating typical ADAS tasks like lane detection in order to achieve real time performance.