Registration is now open for KMWorld 2019. Register now to join us Nov 4 - 7 in Washington, D.C.

Analytics in the IoT public sector: Perfecting with computer vision

This article appears in the issue September/October 2019 [Volume 28, Issue 5]
<< back Page 2 of 3 next >>

Anomaly detection

Anomaly detection offers another cogent example of the ability to increase mission effectiveness and business value, respectively, for the public and private sectors by leveraging computer vision with the IoT. As Shankar indicated, “The gravity of data has always been at the edges: the sources.” This statement redoubles for IoT use cases in which sources are external to conventional enterprise perimeter security, increasing their vulnerability. Organizations can train computer vision models to recognize the status quo of virtually any object, from static images of the brain in healthcare to videos of large public spaces in law enforcement. Thus, if an aberration occurs in those images, the appropriate authorities can deal with it.

“There are large cities in the United States that are using computer vision to detect object changes. Let’s say I’ve got a very important building. I want to detect automatically if a backpack or a large package is left outside of that building unattended,” Bennett noted. “If you have cameras recording those entrances, you want to be able to detect the anomalous appearance of something that doesn’t look like it was just put down for a second, but was deliberately left.” The scale, speed, and accuracy with which computer vision can detect anomalies in medical imaging are critical to improving the efficiency of physicians, for instance. Using this mechanism as a starting place to determine healthcare anomalies enables these professionals to treat more patients (better and faster) than they otherwise could without it. In manufacturing, this deployment of computer vision can rapidly uncover product defects in assembly lines, identify precisely where the aberration occurred, and hasten its redressing to curtail additional loss.

Proactive (not predictive) maintenance

Proactive maintenance is another area in which computer vision can drastically improve IoT’s productivity. The incorporation of drones is a particularly fascinating addition to IoT. These devices not only enable real-time data streaming via video feeds, but also are extremely mobile to service a variety of environments. Subsequently, they function as an effective means of issuing proactive maintenance in a manner that likely exceeds that of traditional predictive analytics used for equipment asset management.

“We’ve got some great use cases around drones for utility [company] use where they can scan lines and automatically detect vegetation overgrowth and prioritize that for crews,” Bennett observed. “You can do that for wildlife counting; you can do that for public land. There are all kinds of uses for drones where you don’t have the human bandwidth to monitor large pieces of land or a utility line on your own.” For the enterprise, computer vision can deliver these advantages for equipment asset monitoring in manufacturing, for example, distributed oil processing infrastructure, and several other decentralized use cases. Moreover, it can triage the ensuing maintenance in a manner based on what is actually happening, as opposed to unfounded predictions about what will.

Instant integration, aggregation

The fundamental way organizations can strengthen their mission effectiveness and efficiency is by swiftly aggregating the results of distributed IoT analytics with traditional, centralized data sources. Bennett reflected that in law enforcement, “What the cop in the field wants to do is collect data, collect information, then push that back into the system for deeper analysis back at headquarters.” When deriving low- latency action on the resultant composite analytics (of which computer vision is becoming increasingly viable), it’s indispensable to view the integrated results “in one screen of glass,” Bennett said. Doing so at the velocity and volume of IoT is readily managed by technologies such as data virtualization, which abstracts data—regardless of location—for seamless integrations.

It “subscribes to the principle that you yield to the data gravity being at the sources, but then you connect to them, rather than collecting them, and bring together a virtual view of that data,” Shankar explained. The increased efficiency of this method directly contributes to greater effectiveness. Shankar noted that a federal nuclear facilities organization was able to reduce costs as much as 80% with this approach, resulting in savings of $500,000. Another federal entity servicing 22 million customers across the country increased customer satisfaction while reducing claims processing times “because they could bring together the data across multiple systems,” Shankar said.

<< back Page 2 of 3 next >>

Search KMWorld