Performance Evaluation of OpenCV-based Applications on the Xilinx ZedBoard

Abstract

This project report presents the performance evaluation of executing a number of OpenCV (Open Computer Vision) algorithms on the Xilinx ZedBoard. The latter contains Zynq-7000 FPGA, a System on a Chip (SoC) from Xilinx, which integrates dual-core ARM Cortex A9. The main goal of the project is to compare the execution times of a number of OpenCV-based image processing applications on the ZedBoard running a Linux kernel with the execution times of the same applications on a Personal Computer running Linux operating system. The OpenCV library was selected for evaluation in this context due to its popularity in computer-vision based embedded systems and its availability for Zynq-7000 devices. The version of OpenCV used in this study is purely a software implementation running on a Linux kernel without any hardware acceleration. The comparison results showed that, on average, the PC runs the selected applications 7 times faster than the Zedboard. The latter achieved an average image processing speed of approximately 5 fps, which is sufficient for a number of real-time computer vision applications. These results also suggest that xfOpenCV will provide better performance (i.e. faster execution time) given the hardware acceleration.