Implementation of Convolutional Neural Networks on FPGA for Object Detection

Object detection is a primary task in computer vision applications. Convolutional neural networks (CNN) have emerged as potential technique and aggressively using in object detection and other computer vision applications. However convolutional neural networks (CNN) are more complex and higher computationally intensive and requires more hardware resources. To address this challenge, field programmable gate array (FPGA) offers significant advantages in terms of parallelism and hardware acceleration which makes them more suitable for accelerating CNN algorithms. This work implements a you-only-look-once (YOLO)v2 CNN on Xilinx zynq-7000 soc FPGA and the results show the increase in performance with reduced processing time and less power consumption compared to CPU based system.