The purpose of this paper is to analyse and test the performance of an artificial intelligence accelerator (Google Coral) for computer vision tasks on low cost embedded devices (Raspberry Pi 4). The first part of the thesis will focus on an in-depth study of Google Coral and then continue with the results obtained, and a comparison, in terms of performance, for inference between the Raspberry Pi CPU, Google Coral and MediaPipe, which is a multi-platform framework for performing accelerated machine learning inferences even on common hardware.
Lo scopo di questo elaborato è quello di analizzare e testare le performance di un acceleratore di intelligenza artificiale (Google Coral) per task di computer vision su dispositivi embedded low cost (Raspberry Pi 4). La prima parte della tesi sarà incentrata su un approfondimento della Google Coral per poi proseguire con i risultati ottenuti e un confronto, in termini di performance per l'inferenza, tra la CPU di Raspberry Pi, Google Coral e MediaPipe, un framework multipiattaforma per eseguire inferenze di machine learning accelerate anche su hardware comune.
Analisi e test di soluzioni embedded con acceleratori AI per task di computer vision
PASQUALINI, FEDERICO
2020/2021
Abstract
The purpose of this paper is to analyse and test the performance of an artificial intelligence accelerator (Google Coral) for computer vision tasks on low cost embedded devices (Raspberry Pi 4). The first part of the thesis will focus on an in-depth study of Google Coral and then continue with the results obtained, and a comparison, in terms of performance, for inference between the Raspberry Pi CPU, Google Coral and MediaPipe, which is a multi-platform framework for performing accelerated machine learning inferences even on common hardware.File | Dimensione | Formato | |
---|---|---|---|
Tesi - 1089440 - Federico Pasqualini.pdf
Open Access dal 22/07/2024
Dimensione
8.84 MB
Formato
Adobe PDF
|
8.84 MB | Adobe PDF | Visualizza/Apri |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.12075/428