FPGA based Deep Learning Models for Object Detection and Recognition Comparison of Object Detection Comparison of object detection models using FPGA

Abstract

Real-time object detection and recognition finds extensive applications in diverse fields such as medical applications, security surveillance, and autonomous vehicles. There are many machines and deep learning techniques that are employed for object detection and recognition. The emergence of a convolutional neural network (CNN) has provided a major breakthrough for object detection and recognition. This work also includes the hardware implementation of the same with the help of Xilinx PYNQ Z2 and Intel Movidius Neural Compute Stick (NCS) which are proved to increase the performance of the system proposed. The results are compared based on three deep learning methods: Single Shot Detector (SSD), Faster Region CNN (FRCNN), You Only Look Once (YOLO). The parameters that are considered are frames per second, probability of detection, and time for computation. The results obtained are performing well compared to existing models.