One of the applications where edge computing with artificial intelligence and 4G or 5G connectivity can add significant value is object recognition.
Cameras can be great for monitoring what happens at specific locations, for taking inventory, for tracking how things change or move, how people behave, and even for finding people or lost items.
The thing about taking pictures or streaming video all the time, though, is that it generates tons of data. If all that data needs to be sent over a telecom network to a central computer location or cloud, to be processed, analyzed and possibly stored, it can quickly become rather costly.
Add to that the fact that the internet was designed to be optimized primarily for downlink rather than uplink data traffic, and it’s clear that large deployments of camera-assisted object recognition are prohibitively resource-intensive.
Now, with edge AI, all this becomes more feasible. When running close to the edge device, ie. the camera, artificial intelligence algorithms can process the image data immediately and throw more than 99 percent of it away.
Only keep what counts
You know, if the purpose is to count people in a queue, all we really need is a number – not the pictures. If the purpose is to detect where a traffic sign has been moved during road works, we only need GPS coordinates once it’s found – not the video stream while trying to find it.
The amount of uplink data applications is expected to grow exponentially. When fast actuation based on real-time data transfer is required, or when large numbers of edge devices are connected to the same use case, 4G or even 5G connectivity may become necessary for flexibility and speed.
Our RAIN platform makes it possible – and in fact very easy – to deploy that technology stack. It provides a horizontal software layer that is present on devices at the edge of the digital network, on cloud servers, as well as on a variety of computing hardware in-between.
Data collection, reduction, analysis and reporting applications all run on top of RAIN and can be accessed through a drag-and-drop user interface. That way, anyone in charge of a use case can easily plan data flows, connect data sources, and create visual, real-time reporting and actuations.
Use cases are only limited by the imagination. And because ours indeed has its limits, we decided to reach out and engage with other innovative minds.
An edge computing contest
Since we are not primarily in the business of developing the AI algorithms ourselves (there are many specialists in numerous domains who do an excellent job at that), we tried a different route.
As part of the UrbanSense 5G Edge Computing innovation project by the City of Helsinki, its Innovation Company Forum Virium, and Helsinki University, we hooked up with Forum Virium and telecom operator Telia to roll out an innovation competition in early 2020 inviting contestants to propose use cases for edge computing.
Together with jurors from Forum Virium and Telia, out of ten candidates we invited and supported three finalists to carry out their proposals during the spring of 2020. The ensuing projects received financial support from UrbanSense. Connectivity and cloud computing infrastructure was provided by Telia Cloud 9, and we made RAIN available for edge-to-cloud software deployments.
Digia is a listed digital networks software and services company with a host of commercial and public-sector customers in Finland. Their idea was to help Stara, the Helsinki City Construction Services unit, detect potholes in Helsinki’s roads by driving around with a connected camera.
Solving real-world problems
One of the findings in Digia’s initial implementation was that the available video software would not allow video frames to be tagged with time and location metadata. In collaboration with Forum Virium and Telia, we were able to engage in a follow-up project in June 2020 to address that particular challenge while bringing the computing capabilities closer to the edge.
By deploying RAIN on an Nvidia Jetson Nano, a small computer for AI applications connected to the dash cam, as well as on Telia’s cloud service, we could quickly install, run and remotely manage an AI algorithm that can recognize potholes from the streaming video. For this use case we then added to RAIN’s library a small extension to add time and location metadata to each video frame that contained a detected pothole.
With the potholes detection case we demonstrated that RAIN can be flexibly used to solve real-world problems and reduce the need to transfer massive amounts of data to the cloud.
The beauty of use cases like this is that every time we add a new application or functionality to RAIN’s library, it can be re-used by anyone on the platform. The system is extremely modular that way. One could easily, say, replace the video input from the dash cam with a Youtube channel, or change “detecting potholes” to “recognizing taxi cabs”, or add an SMS alert at the end of this particular data flow.
Chances are that object recognition could add value to your business, as well. With your knowledge of your industry and our experience with a variety of use cases, surely, together we can find out what makes most sense for you. Book a call with us and let’s have a chat!