Using AI Power

Driving efficiency and effectivity across organizations

• By Stephanie Weagle
• Sep 03, 2020

Video surveillance is commonplace today, but many organizations don’t even realize that they aren’t fully leveraging the video data that their cameras capture. Traditionally, law enforcement and physical security teams use video cameras to monitor areas in real-time and to review footage to glean evidence for post-incident investigation.

Given that staff resources and time are usually limited, it is not realistic to monitor all cameras in real-time, or to manually review all available footage resources post-incident. Even if they have the time, human observations are subject to error or oversight. As a result, most video footage is never viewed or put to practical use, so many organizations miss out on this veritable treasure trove of valuable information.

Progressive organizations have realized that they can, and should, get more value from their video surveillance networks and footage. In recent years, Video Content Analytics software powered by Artificial Intelligence (AI) has emerged as a crucial and complementary technology for video surveillance, because it allows organizations to harness the valuable data in video footage that would otherwise go unused.

Video content analysis allows surveillance security teams to quickly review footage from past incidents, increase situational awareness and response time to evolving situations, and obtain trend data for developing strategies and making data-driven decisions to prevent future problems. The software benefits many industries and is fast becoming a standard part of technology suites, not only for corporate security teams and law enforcement agencies, but also for business groups across organizations.

Driving Agile and Effective Security

Depending on the environment, security and enforcement teams juggle an array of responsibilities, from reducing theft to increasing public safety, or solving crimes. With AI-backed analytics, users can accelerate investigations by searching objects and events of interest with speed and precision.

Operators can filter objects or scenes according to classifications such as male/ female, adult/child, vehicle type, and lighting changes, as well as appearance similarity, face and license plate recognition, color, size, speed, path, direction and dwell time.

This is enabled by AI-driven technology and Deep Neural Network training, which exposes the machine to tagged data to teach it – much like the way a human learns – how to identify objects in video. This enables data to be searched, aggregated and leveraged for triggering alerts. By translating live or archived video into structured data and extracting rich metadata for object extraction, recognition, classification, and indexing activities, video intelligence solutions transform the data into searchable, actionable and quantifiable intelligence for driving investigations, real-time response, and long-term planning.

The ability to forensically filter video based on extensive object classification and recognition empowers the video investigator to pinpoint the most relevant data based on distinct search combinations, such as querying for a person of interest wearing blue jeans and brown coat, heading east between the hours of 4 p.m. to 6 p.m. on a specific date at a particular location.

When such search and filter capability is also extended to the field via mobile technology, officers at the scene of a crime or emergency can quickly search on-site video based on witness descriptions, to jumpstart the investigation before returning to a real-time crime center. Whether in the field or an office, the ability to rapidly search footage across multiple video cameras in a network dramatically decreases the time-to-target and saves hours of investigation and suspect tracking – ultimately preventing crime and freeing up staff to pursue other critical duties.

Improving Situational Awareness with Real-time Alerts

AI-powered video content analytic software is not only for reviewing past events; it also enables organizations to proactively respond to situational changes in an environment, via real-time alerts. Using the same set of object classes and attributes, a video intelligence system can be configured to trigger rule-based, realtime alerts when pre-defined conditions are met. By benchmarking expected activity and by detecting anomalous behavior, users can create alerts for abnormal conditions, such as lighting detected after-hours or a car idling in a pedestrian-only zone.

Video analytics operators can define any number of conditions that require customized alerts- such as crowding and dwelling – for increased situational awareness and proactive and preventative response to a variety of problems.

For example, during the COVID-19 pandemic with its social and physical distancing recommendations, alerting is crucial for detecting and mitigating crowding in facilities of all types. Similarly, dwelling can also be an indication of a problem – whether a medical emergency or an intent to commit a crime – and real-time dwell alerts can be set up to notify when an object or a person has been detected in one spot for an extended duration of time.

Mitigating Risks, Monitoring Compliance

Crowding is a common security and customer experience challenge -– whether in a retail store queue or at an airport security gate – and, therefore, it’s useful to have count-based alerts, which can be configured to trigger whenever the number of objects or people in a particular space exceeds a pre-set threshold. With alerts, operators can proactively detect the early stages of congestion, crowding, or even security breaches when unusual numbers of people are identified in an off-limits area, and quickly assess and preventatively respond to events as they unfold.

One particularly timely example of people counting analytics, is the detection of and alerting for social distancing violations in grocery stores, manufacturing facilities, warehouses and worksites of all varieties. In addition to real-time alerts, managers can also leverage people counting, occupancy and even proximity data to compile reports, dashboards and heat maps for documenting compliance with public health mandates or pinpointing problem hotspots where recommended safety protocols are typically not observed, in order to develop solutions to combat these challenges.

Dashboards and heat maps based on video analytic data can also demonstrate the areas of a private business or public setting that have the highest occupancy and traffic rates – and the peak times of day – to pinpoint where social distancing measures may be difficult to enforce. Municipalities may leverage this comprehensive operational, activity and demographic intelligence to deploy law enforcement to certain city streets or parks where there are high volumes of pedestrians.

Beyond the coronavirus crisis, the ability to detect both patterns and anomalies, empowers organizations to enforce compliance and respond to violations of other important work safety mandates, such as wearing proper safety gear from hard hats to face masks in a work zone. Again, this analytic filter can be used for searching video and triggering alerts; but, over time, the video analytic trend data also can be visualized and analyzed for making intelligent decisions and protecting workers and visitors from everyday hazards.

Face and License Plate Recognition

Often, event prevention and resolution can be accelerated by locating or identifying a specific person or vehicle – whether a criminal suspect, VIP or, in the case of the global pandemic, a selfidenti fied individual who’s contracted the illness. In cases where operators are looking for an identifiable person or vehicle, face recognition and license plate recognition capabilities make searching, alerting on and analyzing video more focused and quick.

“In the wild” face recognition technology relies on watch lists of digital face images to drive identification, video searches and alerts, from watch lists of suspected criminals to those of personnel authorized to enter a sensitive facility. Once a face match is detected, human operators can investigate or evaluate the scene, validate the match and determine how to respond, whether to continue closely monitoring or confronting the individual.

The same principle is true for cars. Law enforcement can, for instance, create watch lists with the plate details of stolen vehicles and trigger alerts whenever a matching plate is detected. Another application is for detecting unauthorized vehicles – especially those associated with previous suspicious or criminal behavior – on a secure premises or in sensitive loading dock areas.

Face recognition has also become a powerful asset for COVID- 19 contact tracing for identifying those who should selfquarantine because they have been exposed to a person infected with the virus.

In a workplace, for instance, an employee can disclose his or her diagnosis to the employer, who can then use facial recognition to identify the employee throughout the work environment over the 14 days prior to the diagnosis. The employer can then identify which other employees or visitors may have had contact with the individual and mitigate further risk by instructing relevant people to self-isolate. This can be done without compromising the anonymity of the infected employee.

Of course, in settings or jurisdictions where there are legal restrictions or physical limitations to using face recognition, it’s helpful to have broader, non-personally identifiable search and alert filters, so operators can apply appearance similarity criteria rather than face recognition – or, in the case of vehicles, license plate recognition.

Distilling Big Data for Operational Intelligence

One of the most significant advantages of video content analytics is that it empowers users to detect not only the granular details – with outstanding precision and speed – but it also can capture and deliver video metadata that has been aggregated over time.

Video content analytics systems provide business intelligence about occupancy, traffic, and dwell patterns. These data visualizations not only help managers identify recurring problems or criteria for expanding real-time alerting and improving response times, but it also drives decision-making by providing accurate insights and trends. Empowered by quantifiable data and trends from video, teams can make better operational decisions based on that actionable intelligence rather than relying on memory or anecdotal observations.

Trend data is important for planning and strategizing how to optimize visitor or customer experiences and business goals.

For example, marketing, operations and security teams in a large event venue or conference center can evaluate historic pedestrian and vehicular traffic to understand where traffic bottlenecks occur, or which entrances are more effective for displaying informational or retail kiosks. In a retail environment, operators can map common customer paths, object interaction, and dwell times. This helps users identify crime hotspots, optimize traffic flow at major traffic interchanges or store locations, track crowd demographics, size and movement patterns; design more effective floor plans or parking lots; and track employee compliance with safety regulations.

To overcome ever evolving challenges, today’s security and operations managers need better technologies for ensuring public and workplace safety and productivity. AI-powered video analytics software drives increased efficiency and effectivity by enhancing surveillance systems most organizations are already using. With flexible architecture options for deploying video analytics in the cloud or on-premises, video analytics technology is more accessible than ever to meet the budgetary, staff and timeline requirements of each individual business. Given that most security organizations already invest in video surveillance, video content analytics is a logical way to maximize that investment with measurable results.

This article originally appeared in the September 2020 issue of Security Today.