Video analytics have come a long way from the motion and object detection analytics that often disappointed early users. From high false alarm rates to a lack of reliable data sets, early video analytics left much to be desired. But the advanced video analytics solutions on the market today are nothing like their past, or even recent, predecessors.
Video analytics have traditionally been viewed as a security luxury: nice to have but limited in form and application. Today the opposite stands true. Advancements in algorithms and artificial intelligence (AI) have propelled the capabilities of video analytics forward to meet a wide variety of use cases.
More than a security add-on, intelligent video analytics have evolved traditional security efforts into profit centers, capable of providing valuable business intelligence, addressing liability concerns and more.
To make the most of next-generation AI-powered video analytics, integrators and other security professionals need to keep their finger on the pulse. As AI, machine learning (ML) and deep learning are constantly advancing, the technology may change quickly. Trends similarly come and go and evolve over time.
However, you can expect current application trends to last well into the foreseeable future as more organizations take advantage of the automated security, safety, operations and additional benefits intelligent video analytics offer.
Know the Technologies
While AI-powered video analytics have the power to transform any video into tailored and actionable intelligence, not all video analytics are created equal. Some analytics come preinstalled on new security cameras while others are available as packaged software solutions; with the latter being more robust.
Enterprise-level software makes existing fixed or mobile security cameras smarter. Software solutions also make it easier to apply new technologies to the analytics. Following are just a few of the advanced technologies found in today’s top-performing video analytic software.
Multiclass Detectors & Descriptor Generators — A multiclass detector describes a single algorithm that can identify more than one type of object in a live or recorded video. Common object types available for detection include people (face or body), vehicles, bags, long guns, etc. Multiclass detectors have an advantage over single class detectors in that they are lightweight, meaning they require less memory for processing. They also improve algorithm testing on the backend, resulting in faster, more precise object identification.
Once an object has been identified, it is then classified to provide even more detail about what is occurring in a video. Advanced classifications do not only take a detection as an input; they also rely on the intermediate results of a deep learning network that is trained to classify based on descriptive features.
Descriptive features can include gender of a person, color of a car, and other factors. In this way, the classifiers can also be lightweight, as well as accurate and flexible.
Re-Identification — Classification by description is nothing new, but smart AI-powered video analytics enable a highly useful process known as re-identification, or Re-ID. Re-ID is a task of detecting an individual across various videos from cameras throughout a facility or campus.
Given a single reference shot of the person of interest, security personnel can retrieve all instances of that person over time from all the connected cameras in the system. Such a function is critical in the case of an emergency physical security breach and could be a lifesaver in the case of a lost child or an abduction, for example.
While facial recognition technology could potentially do the same thing, Re-ID does not evoke the same privacy concerns as it is based on the detection of an entire person, not solely using their face. Re-ID uniquely allows deploying organizations to balance stakeholder privacy and security to build a person-of-interest journey when there is a critical incident. This same technology can similarly be used on vehicles and, hypothetically, on other objects.
New Analytic Types — Today’s video analytics have moved past just real-time analytics to include four other analytic classifications: descriptive, diagnostic, predictive and prescriptive. Each classification fits a variety of use cases for specific points in time. For example, descriptive analytics paint a clear picture of events that have already happened and real-time analytics measure data as it becomes available.
Predictive video analytics use ML to look toward the future, helping predict what may happen based on past events. Prescriptive analytics take predictive analytics one step further to suggest a future course of action and potential impacts for each course of action recommended.
Predictive and prescriptive video analytics are perhaps the most exciting, as they have the potential to automate a variety of processes as well as identify potential risks. Imagine an organization with a high rate of slip and fall events. Predictive analytics could forecast a surge of slip and fall cases in an upcoming rainy season, having recognized a pattern between the time of year and an increase of slip and fall cases recorded.
Prescriptive analytics would then suggest preventative actions such as more frequent floor cleanings, a dedicated space for individuals to dry off after entering the building or deploying more staff to clean areas known to be at high risk for a slip and fall incident.
Both predictive and prescriptive video analytics are an emerging science, requiring vast video data inputs to succeed. While not quite mainstream yet, these types of video analytics represent what is possible, potentially lowering costs, labor and risks for deploying organizations.
Video analytics are commonly marketed as a solution solely addressing security and safety. And although video analytics make incredible safety and security solutions, these functions are already well understood within the industry (i.e. motion detection, facial recognition, person detection, etc.).
Justifying the cost of intelligent video analytics to upper management using strictly security-based benefits has been a longstanding challenge for security professionals. Unfortunately, many organizations and executives don’t see the value of protecting their assets from incidents that have not yet occurred.
To combat this, security professionals are turning to a term frequently used by cybersecurity experts to communicate the value of the solutions they offer — return on security investment, or ROSI. ROSI goes much further than the usual discussion of protection that physical security professionals rely on to make the internal sale to management.
Although the financial investment may be first and foremost in security, the benefits from that investment now extend to all business operations departments across the enterprise.
Business Intelligence — AI-powered video analytics also have the potential to deliver invaluable insights that would be otherwise impossible to gather at the same speed, efficiency and accuracy using manual methods. These insights help guide internal decision-making using facts, rather than inference, to easily identify a best course of action. Not only does this reduce labor costs in the long run and improve employee effectiveness, but also provides a strong ROSI.
Modern enterprises can take this data gathering one step further and feed it to other parts of their organization via application programming interfaces (APIs). For example, analytics from video surveillance systems can provide insights on usage of a kitchen or other nonwork area to inform architectural blueprint software being used for planning construction of a new building. Advanced video analytics platforms also accommodate live video feeds from smartphones and common web browsers, further extending their application versatility.
Compliance and Liability — As business intelligence improves an organization’s bottom line by increasing efficiencies and boosting profits, video analytics also prevent financial loss. Compliance violations and lawsuits can cost companies billions, not to mention lost profits due to a damaged reputation. Video analytics help ensure compliance with federal, local and industry-specific mandates while also protecting organizations, and their CEOs, from damaging lawsuits.
Slip and fall and other workplace injury lawsuits leveled against organizations are unfortunately not uncommon. While video evidence can help refute company liability in such an event, it is often not enough. Video analytics can prevent incidents before they get to the point of litigation, before they occur, by proactively identifying an employee engaging in prohibited behavior (i.e. not wearing the proper protective gear or entering restricted areas).
These same behaviors can also result in a costly compliance infraction from OSHA or other enforcing entities. Detecting these events automatically and remediating behavior produces cost savings in real dollars.
There has also been a recent push across all industries for personal accountability regarding damaging physical security events — so much so that CEOs are being found personally liable for such events, facing jail time and costly fines in many cases. The ROSI offered by security analytics in this case offers not only protection for the organization during a physical security event, but also personal protections from expensive lawsuits.
Balancing Privacy & Progress
Today’s video analytics are not so much a threat to privacy but rather a function of it. This shift largely has to do with a change in marketing and deployment options available to the organization. Integrators are doing their part to bring facial recognition and video analytics out of the shadows, marketing such innovations in the public sphere in ways that are easily understood.
Deploying organizations also have more control over the technologies they recommend, sell and install now more than ever. If company stakeholders are not comfortable with the use of facial recognition in their video surveillance, they can simply turn it off while still taking advantage of available analytics that work based on body signatures.
Responsible analytic software providers are also conscious about the privacy of user data, even using AI to improve privacy and diminish bias. The use of synthetic training data, or general adversarial networks (GANs), is the preferred method of facial recognition training for professionals.
Instead of utilizing a database of photos skimmed off websites for training, GANs use AI to create computer-generated faces to train the facial recognition system. In this way, the images are diverse in skin tone, eye shape, and facial structure to diminish bias and protect personal privacy.
Keeping up with the state of video analytics always requires looking toward the future. In this crowded, fast-paced industry the technology and applications showcased are constantly progressing. No longer a luxury, video analytics are a necessity for any organization looking to increase security, improve operations, and boost their bottom line.
Selecting a thoughtful application of intelligent video analytics from a reputable software provider reassures security professionals that they will not get left behind.
Bruce Ericson is Vice President Global Sales for Vintra.