Analytics in the IoT public sector: Perfecting with computer vision
No matter how complex private-sector deployments of cognitive computing, the Internet of Things, or advanced analytics may be, the goals are still relatively simple. The objectives in business are to maximize profits and minimize loss. Every organization leverages these data-driven processes for competitive advantage; winners are distinguished from losers by the ingenuity of their use cases, not technology.
Not surprisingly, some of the most innovative applications of these developments are in the public sector, which, according to SAS director of Global Government Steve Bennett, is actuated by a different set of aims that can help private organizations achieve their objectives.
“Analytics in government, at the highest level, can be used in two ways,” Bennett revealed. “One is to improve the ability to accomplish a mission. Second is the ability to accomplish the mission more efficiently than you could before. So, mission effectiveness and mission efficiency.”
Since governmental entities aren’t burdened by the financial constraints of private organizations, they have considerably greater liberty to devise new use cases for increasing the efficacy and efficiency of advanced analytics. The most convincing of these deployments broaden the scope and yield of the IoT via computer vision to reduce costs, expand the capabilities of human employees, and better equip organizations for the decentralized data landscape’s challenges.
These same attributes are perfect for maximizing profits and minimizing loss and are therefore as applicable to the private sector as they are to the public sector.
Deconstructing computer vision
The IoT is the most emblematic representation—if not one of the principal causes—of the heterogeneous datasphere that’s scaled horizontally to deluge organizations with a plethora of sources, data structures, and formats. According to Ravi Shankar, chief marketing officer at Denodo, “We started off with the databases, now we have expanded into unstructured sources like streaming data coming in, and then we went into Hadoop NoSQL data repositories. Then came the cloud. Now it’s the Internet of Things where the devices are generating a lot of data.” IoT shifted the data ecosystem from internal to external data sources, many of which are producing too much data, too quickly, for humans to effectively act on. Computer vision addresses this problem in multiple ways as follows:
Object detection: Object detection denotes the ability to identify objects—both holistically as well as different parts of objects—within images. Some objects are exceedingly vast, such as landscapes patrolled by drones. Object detection is one of the primary attributes of computer vision, is responsible for facial recognition, and is the foundation of this technology’s utility for situations in which there is “any kind of large area monitoring that you need to do, and you don’t have enough people to do it,” Bennett said.
Neural networks: Neural networks are the underlying machine-learning models responsible for training detection, identification, and classification techniques upon which computer vision depends. The intricate, multi-layered approach of these models is necessary for the immense detail involved in computer vision for recognizing minute facets of a specific object type.
Deep learning: The tremendous compute power of deep learning is critical for the exorbitant training datasets required for computer vision, which typically works by simply giving such systems a surplus of annotated image data as the basis of its detection capacity. Deep learning is influential for working with neural networks at scale and is responsible for automating aspects of the feature detection that machine learning utilizes.
Surveillance and authentication
Perhaps the foremost use case for increasing mission effectiveness and efficiency (for public and private organizations) is deploying computer vision for authentication. Using this technology for secure authentication directly decreases the incidence and degree of loss, including any associated costs. Computer vision is especially valuable in the distributed IoT settings that are intrinsically less secure than centralized ones. Mihir Shah, StorCentric CEO, described a Department of Defense use case in which “there’s an outpost where there isn’t good connectivity on folks coming in and out of certain organizations, [such as] local folks that they’ve contracted with that need to get identified and allowed on a U.S. base.”
In this instance, facial recognition enables the government to quickly identify those people, look at their IDs, look at the picture that they take, and then match it to what they have on record, Shah divulged. This type of use case is applicable for securing any IoT environment or assets in distributed settings, including private sector ones such as video footage on cruise ships, for example, in which companies are “recording surveillance video, and then once it gets back to dock, it goes into their private cloud for future use,” Shah said. Computer vision could enhance this use case by identifying criminal activity in near real-time to help protect customers on internationally neutral waters, which could preserve organizations’ reputability and mitigate any costly litigation against companies.