Using Google CloudIoT Core in Camera Monitoring for Remote Industrial Operations

Remote Industrial Sites present unique challenges to operate efficiently and effectively. Examples of these sites include construction sites, oil rigs, unmanned industrial stations, industrial farms, and weather monitoring stations. The utility power source is either non-existent or available at a premium, WiFi is not guaranteed for 24x7 Internet connection; and you have to contend with the elements of extreme weather. In addition it can be expensive to permanently station human operators at these sites; and every physical visit to the site for installation, maintenance and supervision adds to the cost of performing industrial operations. On the other hand, it is also expensive to deploy a human operator for remote and continuous video surveillance.

In this virtual meet up, we present the working prototype of a solution designed to address the challenges of Remote Industrial Operations. It is essentially a standalone AI Camera System that relies on solar and battery power to eliminate dependence on utility power. For Internet Connection, it uses cellular wireless LTE CAT M1 IoT connection. The publish / subscribe protocol MQTT is used to carry messages back and forth between the Camera and the GCP Cloud IoT Core backend.

OpenCV is implemented on the Camera device for preprocessing of the captured images, and for functions like motion detection. At the Cloud backend, Cloud Functions are implemented to trigger functions, such as, Image Inference, Object Identification, Device Management, Email Notifications, Remote Control of the Camera, or any other function specific to an industrial application.

The TensorFlow model for a specific field use case is trained and hosted in the Cloud for inference. This AI model can be used for Image Classification, Object Identification and Detection. Optionally, the trained TensorFlow Lite model can be hosted on the Edge device, that, in this case is the AI Camera. As a further extension, the AI Camera System can be adapted for a continuous AI Operations Service Model, that is, with continued usage for a specific industrial application, the model is trained over a period of time and incremental updates applied to the Inference Engine to continuously improve or change its expected performance.

Implementation of this AI Camera System presents an opportunity to effectively drive down the cost of operating remote industrial sites, by automating the remote monitoring of the status and health of industrial operations in real time, reducing site visits, eliminating the need for constant video stream, and cutting down the cost of network connections.

Presenters Bio:

Neil is a Customer Engineer at Google serving clients in the DFW metroplex. Neil has responsibility for technical architecture and solution design for clients examining Google Cloud Platform.

You can reach him at his Linkedin profile:
https://www.linkedin.com/in/kolban/

Iqbal is an IoT Edge Systems Engineer and Architect at US Purtek LLC; his current focus is to develop Cloud IoT Edge based scalable solutions for latency sensitive Industrial IoT Edge Computing applications, with mobile/fixed, wireless/wired, Industrial IoT Edge Computing devices, and ML / AI workloads.

He is a Registered Professional Engineer (P.E.) - Electronics and Communications Engineering- in the State of Texas.

He has previously held engineering and project lead positions with Alcatel-Lucent (now Nokia), Kodiak Networks (now Motorola), and HP Enterprise; and has developed and deployed 3G, 4G cellular wireless network infrastructures for T-Mobile, AT&T, and Verizon.

You can reach him at his Linkedin profile: https://www.linkedin.com/in/k5isj/