Evaluation of the energy viability of smart IoT sensors using TinyML for computer vision applications: A case study

TinyML technology emerges from the intersection of Machine Learning, Embedded Systems, and Internet of Things (IoT), and presents itself as a solution for various IoT fields. For this technology to be successfully applied to embedded devices, it is essential that these devices have adequate energy efficiency. To demonstrate the viability of TinyML technology on embedded devices, field re- search and real experiments were conducted. An embedded system was installed in a turnstile of a Federal Institute, in which a TinyML computer vision model for people detection was implemented. The device counts the number of people, analyzes the battery level, and sends data in real-time to the cloud. The prototype showed promising results, and studies were conducted with a lithium battery and three in series. In these experiments, voltage consumption was analyzed every hour, and the results were presented through graphs. The camera sensor prototype had a consumption of 1.25 volts/hour, while the prototype without the camera sensor showed a longer-lasting consumption of 0.93 volts/hour. This field research will contribute to the advancement of applications and studies related to TinyML in conjunction with IoT and computer vision.