Visual eyes: boosting your IoT strategy with smart cameras and Edge AI

What makes the combination of cameras and edge AI so compelling? Why it’s happening now? and how to start planning your camera and Edge AI project

Boosting your IoT strategy with smart cameras and Edge AI

An engineer prepares to enter the red zone in an oil and gas refinery. A leak of hydrogen sulphide gas in this zone could be fatal. The security camera at the entrance checks that the engineer is wearing emergency breathing apparatus before releasing the door lock.

This is camera and Edge AI in action.

Without built-in intelligence, hundreds of cameras in the refinery would stream video to banks of monitors for an operator to check. Or they could stream video to the cloud for analysis, creating a significant and costly load on bandwidth, network infrastructure and cloud compute. Neither option provides the reliable, real-time response required for safe and efficient operations.

But, together the Internet of Things (IoT) and Edge AI are transforming the commercial and industrial landscape. According to Tractica, AI edge device shipments are expected to increase from 161.4 million units in 2018 to 2.6 billion units by 2025.

Bringing AI to the edge opens up a world of possibilities for companies looking to develop IoT strategies and business models that align with their customers’ needs. AI-enabled cameras, in particular, have enormous potential to transform products and services.

But knowing how to take an IoT idea from proof of concept to prototype and into production is a journey fraught with challenges.

This article describes what makes the combination of cameras and edge AI so compelling, why it’s happening now, and how to start planning your camera and edge AI project.

What do we mean by the ‘Edge’ for AI?

The edge is the physical location where things and people connect with the networked, digital world. An IoT device or sensor is also called an endpoint.

Gartner defines edge computing as ‘part of a distributed computing topology where information processing is located close to the edge, where things and people produce or consume that information’.

The edge consists of many layers at which processing and AI can be deployed. Endpoints and network resources (routers, gateways, servers and micro datacentres) are all candidates for edge computing.

Other terms that overlap with edge computing are fog and mist computing:

  • Fog computing refers to the use of network nodes to perform data, compute, storage and application services
  • Mist computing is the use of compute power on endpoints
  • With Edge AI, machine learning (ML) algorithms are processed locally on an edge device rather than in the cloud or remote data centre. By processing the data as it’s streamed from the sensors, Edge AI can deliver real-time, actionable results.

Deep learning models (such as the Convolutional Neural Networks used for vision analytics) are becoming more capable in edge devices. Having more compute power at the edge means that a model and its inference capabilities can be run both in the cloud and closer to the camera. The architecture of the solution has more flexibility in where processing occurs and therefore what kind of data needs to cross to the cloud. Meta data generated at the edge is returned to the cloud facilitating intelligent monitoring and model refinement.

What’s driving AI to the edge?

Machine learning – and deep learning in particular – requires intensive compute power. Running ML in remote data centres or the cloud has – until recently – been the only option.

But the proliferation of IoT devices and the need for real-time responses have created a number of challenges for this centralized model.

Massive data

Billions of connected IoT devices create a massive amount of data, even if each one only generates a small amount. It’s estimated that security cameras create about 2,500 petabytes of data globally each day.

Sending this data to the cloud for storage and processing is often impractical, inefficient and expensive – especially if much of the data is irrelevant (for example, security footage when nothing is happening).

Edge AI can dramatically reduce the amount of data sent to the cloud, significantly reducing operating costs and the total cost of ownership of the devices.


Many use cases need real-time responses. Examples include:

  • Mission-critical applications, such as autonomous vehicles
  • Quality control systems that check for faults on production lines
  • Safety-critical systems, such as monitoring pipe integrity in a refinery

Waiting for a response from cloud analytics can introduce unacceptable delays. Edge AI eliminates the latency introduced by a round trip to the cloud.


Strict regulations govern the use and transfer of personal data (including biometric and image data). This creates complexity, particularly if data crosses country boundaries. Customers want companies to treat their data correctly, and the publicity from privacy breaches can damage brands.

With edge AI, personal data is kept close to the source, reducing privacy breaches and simplifying compliance with regulations.

Keeping personal data private (and not sending it to the cloud) creates more opportunities for personalised learning applications.


Connecting industrial machines or legacy equipment to the internet can introduce vulnerabilities to hacking. Processing the data on or near the devices reduces these risks.

Limiting the data sent to the cloud can reduce the chance of cybersecurity attacks on data in transit. Intelligent edge nodes can determine the appropriate security mechanisms to use.

However, adding processing nodes increases the attack surfaces – all of which must be protected.



IoT devices can’t always be connected (for example, if they’re located in moving vehicles, remote regions or areas with poor connectivity). A dropped connection may be inconvenient or it may be unacceptable (for example, a door unlocking system must always work).

Edge AI enables devices to be autonomous, operating reliably in remote areas even if they aren’t connected to the cloud.


The move to edge AI changes how costs are distributed.

Smart sensors cost more to manufacture and use more power to operate than dumb devices. However, reduced data and processing in the cloud saves on bandwidth, storage, compute and connectivity costs. Streamlined datasets simplify cloud operations, reducing the need to scale resources up and down.

Although edge devices require additional capabilities to run AI, the cost of smart chips and device hardware is dropping.


Edge AI supports the drive to sustainability, as the power-efficiency of devices improves and data transfer is dramatically reduced.

Edge AI also enables many use cases that support a greener, more eco-friendly approach. Examples include smart cities and energy-efficient buildings, improved manufacturing yields, and greater efficiencies from predictive and proactive maintenance.

Camera and edge AI technology

Many recent technology developments enable viable use cases for cameras and edge AI:

  • High-performance, energy-efficient processors that can operate within the power, thermal and size constraints of IoT devices
  • Configurable, scalable solutions with adaptable domain-specific architectures, using CPU, GPU, TPU or DSP as required for the particular application
  • On-device neural network accelerators and in-memory computing capability
  • AI-enabled camera hardware and off-the-shelf models
  • Frameworks and tools, such as TensorFlow Lite and Amazon DeepLens, that are designed for inference on constrained devices
  • Cloud platforms that aggregate metadata from edge devices and orchestrate their provisioning and maintenance

Partitioning – what should be done at the edge?

Cloud and edge AI can work in partnership. The cloud is best for handling big data, training neural network models, orchestration, and running inferences that are complex, depend on off-device data or can be done offline. Edge AI is best for inference on smaller, self-contained models where autonomy or a fast response is required.

For optimum performance, you may choose to deploy ML on several resources: cloud, edge nodes and endpoints. Unless edge devices need to be completely autonomous, a hybrid model may be the best solution.

To determine where to deploy ML, you need to understand your requirements for autonomy and latency. You also need to identify the data that must be transferred to the cloud for analysis, and the data that can be processed with the available compute power on edge devices.

Your project requirements determine how you should partition the AI in a way that both meets the constraints of the edge devices and achieves the objectives of your project.

Identifying what matters for your Edge AI project

You need to understand how your requirements fit with the constraints of edge AI. For example:

  • Accuracy – models need to be optimized to run in constrained environments, which can result in reduced accuracy but faster predictions. You need to understand what trade-offs are acceptable.
  • Response times – the speed with which you need to serve predictions determines how close to the device you need the ML.
  • Autonomy – you need to determine if the device can – or must – work independently of a central system.
  • Data processing – ML models in the cloud often process data in batches, while ML at the edge processes data in real-time. This can influence the design and architecture of your models.
  • Training – models can learn and improve from the data they process. For example, initially a health monitoring application may detect if a person is wearing a mask. Later, it may learn how to check that the mask is being worn correctly. You need to collect data for continuous improvement.
  • Model updates – the likely frequency and complexity of updating distributed models can impact the choice of architecture and tools.
  • Auditing and explainability – regulations on auditing and explaining decisions made by your AI can impact the information you need to store about ML models and the decisions they make.
  • Privacy – collecting personal image data can create privacy concerns. You need to identify how you will address these and comply with regulations.
  • Security – you need to identify security requirements (for example, encryption, identity and access management) for all edge AI devices.

Preparing to deploy Edge AI

As with any ML project, deployment into a production environment requires planning and preparation (for example, changing the model language, containerization, scaling up).

For Edge AI, there’s also the need to ensure that models are small enough to run in the edge devices. ML models are usually trained in the cloud with as much compute power as they need, using 32 bit floating point precision format for maximum accuracy. The resulting models – possibly hundreds of megabytes in size – need to be reduced to tens of megabytes to run in edge devices.

A framework such as TensorFlow Lite is designed for conducting inference on IoT and mobile devices. Models can be optimized for this with techniques like quantization and optimization libraries such as NXP Glow.

Quantization uses lower precision formats to reduce the size of the model, which can result in reduced accuracy. If accuracy loss isn’t acceptable, there are ways to compensate and other methods of optimization.


Managing Edge AI in operation

Edge AI brings new challenges for DevOps – managing ML models in possibly thousands of edge devices. This requires:

  • Developing a delivery model for managing devices remotely, including deploying, monitoring and fine-tuning models, and tracking configuration and updates
  • Collecting, storing and processing data that isn’t sent to the cloud immediately, but may be required for on-going training or archiving
  • Handling failures (for example, if the model can’t make a prediction or gets it wrong, a user may be able to supply the correct answer)
  • Monitoring the health of the hardware

What can cameras and Edge AI do?

Many applications for cameras and Edge AI already exist, and many more will become viable as technologies mature. Sensor fusion (combining data from multiple sensors of the same or different types) enables further innovation and development opportunities by providing a more accurate world model.

Smart vehicles

  • Fully autonomous vehicles are in an AI category of their own.
  • But there are other ways smart cameras can improve safety in vehicles.
  • For example, cameras can analyse road conditions and driver behaviour (such as falling asleep) to alert the driver and reduce accident rates.
  • Cameras on vehicles can also assist with safe manoeuvring and parking.

Retail and marketing

  • Camera using AI can track shopper behaviour to help the retailer improve product placement and store lay out.
  • Systems to detect footfall and people counting can assist with customer service, queuing and product placement.
  • Cameras play a key role in checkout-free stores.

Smart homes and devices

  • Facial recognition can support home security, for example, allowing access to family members and warning of intruders.
  • Mood recognition enables smart devices to tailor services, such as creating mood-appropriate playlists.

Smart cities

  • Cameras installed in street lights can detect people and vehicles, helping to improve safety and security.
  • Traffic lights can be synchronized in response to traffic conditions, or to help responders travel quickly in an emergency.
  • Cameras in car parks can direct motorists to free spaces, and warn of behaviour such as a person lingering beside a vehicle.

Workplace safety and security

  • Facial recognition can enable security measures, such as controlling access to buildings or restricted zones.
  • Cameras – in conjunction with other sensors – can identify trip hazards, slippery surfaces and leaks, and track the movements of cranes and other industrial vehicles.
  • In a pandemic, cameras can check mask-wearing, social distancing and even hand washing practices. Thermal cameras can highlight raised temperatures, and audio from cameras can detect bouts of coughing.
  • Cameras in smoke detectors can identify the location and severity of a fire, enabling targeted use of sprinklers and providing detailed information to emergency responders.


  • Cameras installed in factories can detect flaws in products as they’re manufactured or assembled, increasing yield and reducing waste.

Remote inspections and rescues

  • Drones with cameras can inspect remote or dangerous sites, such as oil rigs. They can aid search and rescue missions, helping to locate stranded people and mapping the best route to reach them.

A vision for the future

Cameras and edge AI can deliver real value to companies that see its potential and can address the challenges it brings. Steadily-maturing Edge AI and camera technologies, sensor fusion and 5G are driving the opportunities for successful and innovative use cases.

But achieving ROI isn’t easy. From proof of concept and prototype to development, deployment and operations, there are many pitfalls for the unwary.
Success comes from partnering with experts, who can help you navigate through the difficult decisions and obstacles ahead. They can help you:

  • Validate use cases, develop proof of concept and prototypes, plan your roadmap
  • Choose the technology stack that best meets your requirements, integrate products and develop applications
  • Implement and provide a delivery model that is flexible and scalable so you can get to market faster

Is Edge for you? Get in touch with us

Consult Red is a technology consulting company helping clients deliver connected devices and systems, supporting them through the entire development journey.

Contact us to talk about achieving value with your camera and Edge AI project.