Multi-hand finger counting hardware/software system with Xillybus IP core on Xilinx Zynq

Abstract

As designers push the capabilities of embedded applications, hardware acceleration in embedded systems has become a topic of interest. Embedded reconfigurable systems display promising advantages: higher performance while drawing lower power with better reliability and scalability. Even though research efforts have showcased the efficiency of accelerating computationally intensive tasks in hardware, mainstream adoption has been slow, mainly due to the difficulty of designing an effective system to harness the full potential speedup. Avenues to ease the development process are now available in the form abstraction at multiple levels: high-level synthesis, communication, operating system etc. In this project, we utilize the Xillybus IP Core to simplify communication between the hardware/software system of Xilinx Zynq. We implemented a hardware accelerated multi-hand finger counting application and observed ~25% speedup for accelerated portions. GTK+ library was used for GUI development, and OpenCV library was used for image processing and geometric data structures. Besides detailing the performance increase in our report, we also showcase how high-level synthesis, communication and operating system abstraction eased the hardware design process, allowing us to harness strengths of both a general processor as well as reconfigurable fabric more efficiently when designing embedded applications.