Розробка нейронної мережі для розпізнавання людей на апаратній платформі Raspberry Pi

Abstract

The necessity of introducing object recognition algorithms in modern video surveillance systems is shown. Variants of standard neural network architectures and simple solutions for embedded systems are analyzed. A neural network has been developed that allows people to be recognized on the Raspberry PI hard ware platform. The network allows us to process the video stream in near real time. The accuracy of object classification using the developed neural network is 5% lower than that of the well-known YOLO algorithm.