FPGA based object parameter detection for Embedded Vision Application

Abstract

Real-time image/video processing contains complex algorithms, and the number of computation operations such as a number of additions and the number of multiplications, which are solved by CPU, results decrease in processing speed and consume more power. On the other hand, FPGA has high processing speed and low power consumption and it also has the capability to work with CPU. Hence, sharing the load and performing complex tasks, results increase the system performance. In this work, there is the development of IPs for single object detection and canny edge detection using different filters. These IPs can be used in FPGA for real-time image processing. The tool which is used is Vivado HLS, to generate the hardware accelerator and the library which is used is the xfOpenCV library. Both are developed by Xilinx. XfOpenCV library is the optimized version of OpenCV means to reduce the complexity of an algorithm for fast performance. Single object detection detects the object in real-time by using FPGA based system and xfOpenCV APIs. Similarly, edge detection using a different filter shows the effect of filters, which means removing noise in the image and making it smooth, for edge detection. Smoothing image also reduces the number of computations (Addition and Multiplication). Resource utilized by the generated IPs is very less which makes the system less heavy and more reliable.