Performance Evaluation of the Nvidia Jetson Nano through a Real-time Machine Learning Application

he use of applications with Machine Learning (ML) is increasingly frequent and their daily use demands more capacity for processing information in real-time. Many devices deploy their infrastructure in Cloud Computing environments, where latency and network partitions are the main challenges they must face. However, in environments such as Dew Computing, these problems are mitigated since the information processing is carried out directly on the data source, so an Internet connection is not necessary. This article evaluates the performance of the Nvidia Jetson Nano platform, under a Dew Computing approach, with an application that uses ML to solve a problem of identification and counting of land vehicles, in real-time, in the city of Quito, Ecuador. Two types of experiments were conducted, the first experiment aimed to evaluate the use of processing resources (CPU and GPU), device temperature, power consumption, and the amount of RAM used in the ML task. The second experiment sought to evaluate the effectiveness of the platform combined with OpenDataCam.