Unlocking New Possibilities with Embedded Vision Cameras Using Hyperspectral Imaging  

Whether it is industrial applications powered by automated robots operating safely in the same space as humans, consumer applications including smart vacuums simplifying our homelife, or a future world where we all travel from place to place in self-driving cars, teaching machines to “see” opens all our eyes to what the future might hold. 

Powered by AI algorithms, machine vision removes the need for human intervention in visual tasks. It enables computers to analyse and process images more effectively (higher accuracy) and efficiently (faster) than humans, returning actionable insights to automate tasks and decision-making. 

Early machine vision cameras were bulky and expensive devices that transmitted images to external hardware and software for analysis and interpretation. The miniaturisation of processor and imaging technology has led to the proliferation of embedded computer vision systems, introducing more compact smart cameras into many more applications. 

At Living Optics, we believe embedded machine vision can get even more intelligent by upgrading from RGB datasets to higher spectral resolution images. Our mission is to develop the next generation of embedded vision powered by hyperspectral technology. To do that, we need to redefine how visual data is captured and interpreted, designing new, embedded hardware capable of providing hyperspectral insights in real-time at the network edge. 

What are embedded vision cameras? 

An embedded vision camera is a machine vision system that integrates both image capture and image processing into a single device. Rather than capturing images for processing off-device, the computer vision – the actual processing, analysis, and interpretation of the visual data – takes place locally, embedded into the device itself. 

Embedded camera machine vision provides potential benefits compared to typical machine vision, providing lower-cost, more compact, and more responsive systems. You no longer have to transmit data to an external PC or the cloud for analysis. Embedded vision cameras remove this latency, allowing for real-time autonomous decision-making. 

This technology is made possible by improvements in edge computing, with advanced processors that can handle large, embedded computer vision workloads locally, returning quick, intelligent analysis of visual data. 

Embedded vision cameras are regularly found in autonomous vehicles, drones, smart traffic hardware, and other IoT devices. Typically, they are used in systems that only need to perform a small number of targeted tasks. This means the hardware and software can be stripped down to work on-device rather than needing larger-scale compute capabilities that are more common to traditional machine vision. 

Consider the example of a CCTV or video surveillance system. A traditional system would simply capture images for humans to monitor and review for suspicious or unusual behaviours.  

Machine vision could automate this process in a number of different ways, perhaps automatically flagging behaviour that differs from the system’s pre-defined understanding of “normal” or even having built-in facial recognition to trigger when a new person is seen. Machine vision cameras would capture video and transmit it to an external computer for analysis. Embedded computer vision would perform this analysis locally on-device. 

You can think of embedded vision cameras as a subset of the broader machine vision field. While embedded computer devices struggled to provide the processing power needed for machine vision tasks in the past, they have since found use across several fields. 

Embedded vision applications 

Advances in imaging and processor technology mean the performance of embedded vision cameras now approaches the capabilities of larger machine vision systems.  

The level of compute for a given volume has increased significantly over the years. This enables lightweight, lower-power, embedded RGB cameras to be integrated into existing systems, providing new machine vision functionality.  

These embedded camera machine vision systems facilitate imaging tasks that are not possible using larger, more traditional systems.  

• Robotics: enhanced automation, decision-making, and more, improving the performance of both industrial and collaborative robots. In particular, embedded camera systems offer significant benefits in no or low network connectivity environments where traditional machine vision has not previously been feasible. 

• Security: more advanced and affordable video surveillance systems with built-in intelligence to understand objects and scenarios such as perimeter breaches at secure locations. Beyond CCTV style deployments, embedded vision is being incorporated into drones and indoor security robots to monitor and guard critical infrastructure. 

• Healthcare: enabling new imaging techniques for improved diagnostics and a range of patient monitoring solutions from motion analysis during rehabilitation to remote healthcare opportunities. 

• Agriculture: embedded machine vision for precision agriculture, which includes weed detection, improved planting, crop monitoring, and better resource efficiency. 

• Smart traffic management: applications across urban areas to understand real-time traffic conditions, returning data to optimise the flow of people and vehicles and improve safety. Embedded vision cameras can perform various tasks, including traffic flow analysis and incident detection. 

• Automotive: used in combination with other sensors for driver assist systems. Embedded vision cameras for low latency decision-making are critical to developing future semi and fully autonomous vehicles. 

Embedded cameras deliver machine vision in a smaller form factor with local data analysis for faster, more autonomous operations. What if we could supercharge machine vision technology by replacing the data it has to work with something more advanced? 

Enhancing embedded vision with hyperspectral imaging

Spectral imaging technology goes beyond RGB cameras to see more than meets the eye, measuring light using tens or hundreds of spectral bands instead of just three. 

This spectral resolution allows the camera to return a clear picture of light spectra without sacrificing spatial information. It adds new layers of hyperspectral data to see in more detail, simplifying computer vision analysis while also leading to more advanced use cases. 

A typical computer vision system using an RGB camera only has data on shape and simple colour to interpret an image. Hyperspectral imaging technology provides an in-depth understanding of the light spectra across the image to identify materials, segment scenes, accurately track objects across an image, and more. 

With richer imaging data on offer, hyperspectral cameras have the potential to effectively replace traditional RGB cameras to upgrade embedded vision capabilities. The addition of hyperspectral data could transform applications across various sectors. Examples include: 

• Security: surveillance systems that incorporate hyperspectral data to identify materials present (potentially flagging known dangerous substances or signatures) and track objects across images even when partially or intermittently obscured. 

• Medical diagnostics: providing more accurate imaging to improve diagnostics and treatment. Imaging techniques and monitoring systems powered by hyperspectral data can identify tissue types in real-time or estimate blood oxygenation. 

• Precision agriculture: spectral analysis to track detailed plant health metrics, such as chlorophyll levels and nutrient deficiencies, with mobile hyperspectral cameras attached to drones and vehicles, or used as handheld devices. 

• Environmental monitoring: precise spectral imaging data to monitor natural resources, environmental changes, and the effects of human activity. 

• Industrial inspection: detailed and reliable hyperspectral imaging to better understand manufacturing processes, improving quality control and ensuring operations adhere to standards and regulations.  

• Autonomous navigation: Enhancing the capabilities of autonomous vehicles with hyperspectral vision, enabling them to detect and differentiate between different objects and materials for safer and more efficient navigation. 

Hyperspectral cameras are typically complex and fragile devices. Plus, they output significantly more data than RBG alternatives, leading to lower frame rates. To make these use cases a reality, we need technology that maintains the benefits of embedded machine vision while handling new spectral imaging requirements.  

We need to develop a compact, lightweight, sturdy hyperspectral imaging camera that can be embedded into existing systems. We also need new compression techniques to output and process significantly larger datasets in real-time. 

The Living Optics Camera

The Living Optics Camera is a next-generation hyperspectral system capable of delivering high-resolution spectral data at video rates from a compact, light, and easy to use device.  

A field-ready hyperspectral system, the Living Optics camera is also the perfect discovery tool to assess the requirements and potential value of spectral data in a range of new solutions. This includes embedded spectral imaging camera systems for machine vision. 

We are making it our mission to enable the next generation of embedded vision powered by advanced hyperspectral technology. Our technology stack is capable of being packaged as an embedded vision camera, and we expect future iterations of our camera to do just that. 

Get in touch today to become an early adopter and shape the future use cases of embedded vision.