Data has always been a business game changer as a rear view indicator. It’s been defined as the new oil. Most of the time, data gets collected, stored, and then analyzed to find the right insights through multiple sets of tools. With this cumbersome approach, reaching those critical data points requires considerable time. In the process, opportunities are lost and greater costs are accumulated.
And, with more companies across all types of verticals seeing that they can tap into even more data through Internet of Things (IoT) platforms, that means even more data to sift through. So, if data is the new oil, then the question is, who is your refiner? Thanks to AI, solutions are emerging to fix these issues and start providing insights that automated systems can use in real time.
The Video Surveillance Case:
Video surveillance is one extreme example of the massive amount of data being used in industries like transportation, energy, construction, manufacturing and municipalities. Camera equipment has become significantly cheaper and more accepted in workplaces as necessary tools. With video comes visual data from numerous cameras — sometimes hundreds or thousands — and the inability of humans to review the amount of hours of video captured by all those cameras. Companies install video cameras. However, they stop watching the video feeds over time. The data only becomes important when a specific event requires review. Then, video monitoring becomes an exercise in forensics to pinpoint and verify a particular event or anomaly.
According to Khamis Abulgubein, Principal Product Manager, IoT Applications for Nokia, video data is fast becoming a pain point. “Each camera produces 1MB of uplink traffic and storage per second, 24 hours a day, 365 days a year. Of that, only about 1% of is typically relevant data and even less contain data that requires an instant response, including traffic delays, streetlight outages, crime in progress, and other safety concerns. Compare those numbers with a city like Bristol UK who has over 700 cameras and you can extrapolate that it is a big problem.”
He added, “Furthermore, storing this information creates a privacy concern where people and cars are constantly monitored and their information are stored with the rest of information. Ideally, you want to quickly discard the video footage and only keep the insights; along with the minimal amount of footage needed to provide context or only prioritize those cameras for human review at that moment in time.”
Looking to Ease the Pain Points
Verticals like the public sector, transportation, and utilities cannot realistically apply the resources to viewing all the available data from these video cameras. Knowing that there is so much idle time throughout the video data only adds to the belief that the video cameras should be left alone until specific data needs to be located. Yet, companies or city governments need to be alerted to situations that need their attention immediately, such as crime or traffic congestion.
Currently, there are vendors who can record and compile all the video data on a company or organization’s behalf. However, these do not address the pain point of response time or comprehensive insight delivery. In an ideal world, these verticals would be able to have data insights selected in real time and reported to them. That way, they could conserve resources, identify critical areas of improvement more quickly, and leverage the insights to make decisions with impactful results.
Artificial Intelligence and IoT Video Analytics
The ideal world is here thanks to unassisted artificial intelligence (AI). It is driving the emergence of IoT video analytics and turning IP cameras into smart IoT sensors that can help a business scale. To understand how unassisted AI changes everything, Andrew Ng, Co-Founder of Coursera, an AI pioneer at Google and Baidu and thought leader on machine learning, put it this way: “Pretty much anything that a normal person can do in <1 sec, we can now automate with AI” And, this is 95% of what we do everyday.
Less human configuring is needed thanks to the power of unassisted AI. For example, the unassisted AI reviews the video scenes in real time to collect the data that is critical for human review or that requires further analysis. Hours of unnecessary video can be deleted. This reduces the amount of video storage. In remote areas where connectivity may be an issue, network uplink is no longer problematic.
The Case For Unassisted AI
Unassisted AI is one aspect of how artificial intelligence can be applied. According to Marc Jadoul, IoT Market Development Director for Nokia, it’s the unassisted AI component to their Nokia Impact Scene Analytics solution that provides a key differentiating benefit for verticals like transportation, utilities, and the public sector. “Supervised AI can be very beneficial in certain situations. However, for it to work requires thousands of hours of video clips to teach the system. This includes feeding the system any number of situations and outcomes so it can learn how to identify those in the future.”
He added, “Yet, there are certain situations and outcomes where it would be better to avoid — or even impossible — creating these video clips and introducing that data to the system, especially in terms of car accidents or other experiences that may have dangerous outcomes. Instead, unassisted AI gets to work immediately and teaches itself as it goes to deliver a faster scale and a higher success rate. Our Nokia solution uses machine learning to review video feeds continually, translates scenes into easier-to-analyze patterns, and then marks the statistical anomalies it identifies for human review.” The ability to avoid certain situations altogether can further increase the human savings while reducing network usage and enhancing response times.
Smart City Use Case
In a smart city environment, an IoT video analytics platform would survey traffic patterns and crowd flow. The real-time ability could also work together with traffic light sensors to understand the best pattern of traffic light timing. Insights for various parts of the day and night would then direct how to alleviate congestion.
A city could also gain real-time control over energy consumption or get meaningful data for optimizing neighborhoods. For example, this might involve planning where to position street lights tied to where crowds congregate for events. When combined with other IoT sensors in a city, and other applications like smart lighting and smart parking, this platform could also help with shaping infrastructure changes or future development.
There are numerous other situations for automatic anomaly detection. For example, it can detect stalled vehicles, roadway debris, and abandoned objects on a street or placed near crowds. Other venue applications include parks, airports, train stations, parking lots, and amusement parks. Currently, Nokia Impact Scene Analytics is undergoing trials in cities and venues around the world to quantify the benefits.
Future Applications For Unassisted AI and IoT Video Analytics
Commercial implementation across verticals is just beginning to take off as the technology continues to evolve. IoT video analytics is set to become an ideal platform for home surveillance applications. This includes enhancing security and energy usage. Additionally, it could be added to cellular cameras to provide a flexible video monitoring option. The wireless capability can deliver video insights similar to the stationary commercial and residential applications.
Also, unassisted AI can be combined with more IoT devices on a commercial and consumer basis as comfort levels increase and use cases emerge. It’s already been pegged as an integral component to move autonomous vehicles forward. With its ability to learn as the autonomous vehicle environment continues to evolve, it may play an increasing role in solving many of the current issues related to vehicles, human behavior, and city infrastructure.