Hiring Artificial Intelligence to make Edge Smart
In the concept of edge computing, through distributed architecture functionality, resources such as storage, processing or GPU are located closer to the source of data.
Edge servers enables variety of services – especially time sensitive services – to be executed almost immediately. This concept, besides resolving latency challenges, also saves large amount of traffic to be send up to cloud servers. EY survey taken in 2018 indicates that 27% of global telecom companies are implementing/expanding edge computing in 2019*. International Data Corporation (IDC) predicts that by 2025, nearly 45% of the world’s data will move closer to the network edge.
Latency and bandwidth optimizations have clearly became most import ant factors in edge computing importance.
Flexgent IoT solution opens opportunities to enable smart edge without investing large capital in setting up edge servers. It brings intelligence right to the gateway. The easiest way to demonstrate how Flexgent can accomplish this is to dive straight to real example.
Nowadays, with available AI tools and growing popularity of machine learning, we observe variety of object detection applications already deployed into the market. Objects can be detected directly on video streams, that are coming from unique type of IoT sensing devices – cameras. In our scenario gateway is responsible for sending a stream for further processing. If trained neural networks are used to process images and detect objects than it typically requires significant multi core processing power. Such processing capabilities are offered by cloud service providers that are using powerful servers equipped with large numbers of GPUs.
Service providers are now investing to build powerful edge with processing capabilities – however hardware requirements for effective machine learning are high and drive costs up.
Softgnet’s Flexgent provides light weight hardware agnostic micro-services engine software that is a foundation for IoT devices and connectivity management. One of the key engine features is its flexibility. Creating a Phyton based micro service application is sufficient to add new, smart features to IoT gateways. For the sake of this use case we have created a micro service utilizing our AI framework that provides image classification for video streams. Our application uses pre-trained neural network to analyze frames in the stream object detection. Only upon detecting the specified object the microservice sends frames to the cloud for further processing that might require more hardware resources not available at the edge.
Flexgent also provides set of core functions such as micro services life cycle management and telemetry used for IoT device and gateway monitoring.
For this PoC we have used a cost-effective NVIDIA evaluation board called JETSON Nano and a standard web camera. We have simulated management cloud by bringing MQTT broker function to dedicated PC and to simulate data processing cloud we have created a script that displays frames received from the IoT gateway. We have managed to make edge IoT gateway smart so that it only engages cloud when a specified object had been detected. This solution allows us to optimize the edge and limit traffic to the cloud. JETSON Nano with our Image Classifier AI microservice and Flexgent framework allowed us to reach the following performance characteristics:
| Image resolution | FPS |
| 224×224 | 10 |
| 320×320 | 8 |
| 416×416 | 6-7 |
Please let us know if you are interested in making your edge smarter by utilizing Flexgent software to build AI based functions of your IoT gateway. Our experts will be able to help you optimize AI for limited capabilities of the edge, come up with efficient neural network models and connect your IoT to our Flexgnet Management Cloud functionality.