Applications of deep learning and artificial intelligence methods to smart edge devices and stereo cameras

Abstract

The aim of this work is to test a computer vision application that, thanks to edge computing and the use of devices optimised for artificial intelligence, allows the distance of objects from a fixed point to be measured. The distance is calculated between two or more points of interest with a certain accuracy and within a certain range. The point of interest from which to measure the distance is determined by a fixed point or an object recognised through deep learning techniques, with which the neural chip can perform very efficiently and with low energy consumption. In this work we used an Edge AI device with stereo cameras, a Luxonis OAK-D, equipped with an Intel Myriad-X neural chip and an improved version of the open-source Luxonis DepthAI library. The application is very complex and has the potential to be used in a variety of areas where precise positioning and real-time object recognition is required. The areas of applicability can be diverse, including assisted navigation, spatial data analysis, robotics, surveillance, health and safety, construction and engineering, surveying and mapping, transport, sports and fitness.