FPGA accelerated DeepSORT object tracking

Object detection and tracking is one of the most important vision algorithms in intelligent environments, applications based on it are the driving force behind the growing deployment of edge devices. As FPGAs can implement application specific hardware accelerators, they not only offer outstanding performance but also increased energy efficiency which is a very important factor in edge devices and the most recent FPGA families include processing elements which greatly increase the efficiency and performance of AI inference. In this paper we present a heterogeneous FPGA based implementation of the popular DeepSORT tracking algorithm.