1 of 5

smart-camera demo

A proof-of-concept of AI classification running on RockPi4B with Linaro TRS, Docker, TensorFlow, AWS Greengrass

2 of 5

Introduction

smart-camera is a proof of concept putting together technologies that will be further developed in the Software Defined Camera Blueprint project (SDC) involving Linaro & Arm:

“The goal of SDC is to provide a demo-able software stack that allows for the deployment of Smart Camera / Vision use cases. Those essentially involve image capture via a camera and AI processing. A typical use case is object recognition. The architecture of the project allows flexible deployments with workloads running on the edge or in the cloud. This is accomplished via containers.”

3 of 5

smart-camera Overview

Build host

(Linux with Docker)

“make”

Source code &�(trained) AI model

AWS S3�(storage)

AI model

AWS SageMaker

(ML service)

AI model

Compiled model for AArch64 Linux

Docker images packaged as Greengrass components

(capture & inference)

AWS = Amazon Web Services� No strong dependency, could by another cloud� (Alibaba…)

IoT = Internet of Things�ML = Machine Learning�TRS = Linaro Trusted Reference Stack

RockPi4B�running TRS

AWS Greengrass

(IoT service)

Greengrass components

(application software)

Compiled model

Monitoring / control for software deployment

AWS EC2�(compute)

(optional)

Inference

Inference

Capture

4 of 5

Demo!

Video, 26m26s, link

5 of 5

Thank you