Acceleration of object recognition algorithm on embedded platform

This paper deals with the new trend of algorithm development in embedded systems, more specifically, object localization and classification algorithms. Different one-stage and two-stage detection algorithms are presented, which have different characteristics. The most relevant algorithm on which the article will focus is YOL0. The YOLOv3, YOLOv4, and YOLO Fastest versions are studied, as well as different acceleration libraries such as PyArmNN and NNPack. The embedded platform on which we work is the Raspberry Pi 4, performing performance and accuracy tests with different library and hardware configurations