Fujitsu’s O-RAN virtual Central Unit/Distributed Unit (vCU/DU), featured in a recent blog, brings the power and potential of AI to the O-RAN edge using NVIDIA’s converged accelerator platform, paving the way for all manner of potential enterprise applications. One example is Actlyzer, an AI-based video analytics application for real time human behavior recognition, which Fujitsu is in the early stages of commercializing.
This is an early example of how it is possible to deploy AI applications as an integrated part of the Fujitsu O-RAN Edge, and enable use-case driven enterprise applications (such as Actlyzer) by reusing the compute and NVIDIA GPU.
How Actlyzer cuts behavior-recognition implementation time, effort, and cost
The crux of what makes AI work is the modeling and training that happens before an AI application is implemented, which involves a serious amount of upfront effort. 77% of overall AI development time is spent preparing the necessary data, and 20% is spent building the deep learning models that enable an AI to function. Here’s the problem – business owners and enterprises can’t do that themselves. They need specialized data scientists to do it who are familiar with building AI models. That’s not expertise that is generic or easily available off the shelf, and generally the people who are qualified to do it don’t spend their time building AIs to generate business outcomes. Now, the Actlyzer real time video analytics system enables business owners to implement behavior recognition AI with no specialized expertise.
What Actlyzer does is prepopulate the necessary AI learning models in an easily reusable form, so that in essence, the Actlyzer system comes “pre-trained” to detect “building blocks” of human behavior in video feeds. These building blocks are essentially behavior ingredients known as “basic actions,” which can be whole-body actions as sitting, walking or running; or they can be body-part actions such as raising the arm, tilting the head, and even holding specific objects in the hand. Actlyzer can also recognize facial expressions and personal characteristics including height or styles of dress (such as a uniform or a face mask).
Actlyzer can be programmed to recognize combinations and sequences of the basic actions and to take specific actions as a result. An example might be a person in a clothes store who is looking up at items on a high shelf, or who picks an item off a hanger and starts looking around—in which case Actlyzer could recognize these combinations as a customer looking for assistance and alert staff. Another example might be checking the sequence in which assembly-line workers are performing tasks to help ensure work is completed in the correct order, or simply monitoring the time taken to perform tasks. Actlyzer’s ability to recognize specific personal characteristics could even be a valuable aid in locating a lost or unaccompanied child at a festival or mall.
Deploying networked AI video analytics without specialized skills
The other piece of the puzzle is that Actlyzer allows a nontechnical user to assemble these basic actions using a drag-and-drop graphical workbench that requires only minimal technical skill to operate. The result is a system that’s highly accessible to nonscientific, nontechnical people.
There are wide-ranging possibilities for Actlyzer. In industrial or manufacturing settings, for instance, the system is capable of monitoring many safety and performance aspects of worker behavior, from safe lifting practices to whether or not protective gear is being worn to whether or not a commercial driver is wearing a seat belt. The current system is prepopulated with over 100 basic actions and more are likely to be added as the system matures. Other ideas that have floated include configuring it to recognize an active shooter, detect in-store theft, spot people entering protected areas (think “keep off the grass” at a park), or alert staff if people tamper with displays or exhibits.
How service providers could offer Actlyzer’s AI video analytics as a service on their networks
Actlyzer requires network connectivity and can be run in a variety of settings, including at the edge of the 5G O-RAN, in which case the system is run on the Fujitsu AI-enabled vCU/DU utilizing the compute resources of NVIDIA’s converged accelerator.
For service providers deploying 5G networks, this is just one example of a range of potential applications that could monetize processing power if offered as value-added services for their enterprise customers. Learn more about Actlyzer and how it’s already being used in real life.