Page 72 - Security Today, July/August 2021

“While the targeted violent attacks on Houses of Worship are what make the news, they face a range of incidents from loitering, graffiti, and vandalism, all the way to life- threatening violence. ”
signed to assist with this task. Powerful AI search technology can detect objects and filter on attributes such as clothing or vehicle color, direction of movement and regions crossed. Some solu- tions even have the ability to track a single person across mul- tiple cameras. In addition, advanced technology to “squeeze” long hours of video into shorter clips also reduces manual review times, often by 10 times.
Implementation of the technology is also important. Some use cases may require on-premise installations but many of to- day’s offerings take advantage of cloud computing to reduce costs, simplify deployment, and provide flexible, up-to-date mod- els. Analytics solutions that can leverage existing infrastructure help reduce the costs of implementation, enabling the technol- ogy’s adoption in more locations.
FACIAL RECOGNITION
Facial recognition technology is a powerful surveillance tool whose popularity appeared to wane somewhat in mid-2020 as tech companies such as Microsoft, IBM and Amazon put a moratorium on selling their software to law enforcement. The technology continues to advance, however, and there’s been a re- surgence of interest in adoption. While there are still concerns around facial recognition, such as using technology to confirm an identity of a person – facial detection continues its adoption as it only strives to recognize an object as a face in an image.
Despite recent developments in machine learning, face recog- nition is challenging due to the great variability in head rotation and tilt, lighting intensity, facial expression, aging, etc. It is some- what surprising that today, the main challenges for automatic recognition remain the same as with those identified 52 years ago by Woody Bledsoe and coincide with the problem of PIES (pose- illumination-expression-structuring) in face recognition.2
Also, until recently the diversity of the samples used to train the AI models has been lacking, with white males being falsely matched less frequently than BIPOC faces and women3. Greater awareness and expansion of training data will help in correcting this bias.
Many Houses of Worship continue to explore implementing face recognition and detection, especially in conjunction with other security and access controls methods.
WEAPONS DETECTION
“The biggest priority is the threat of weapons, so we’re heavily focusing on weapons detection,” Suthar said. “Many of their in- ternational locations have metal detectors, with some larger loca- tions disallowing even mobile phones. One of the challenges of deploying these universally is the requirement to force people to use only designated doors and entrances.
“Some mandirs have grand entrances which would have to be permanently closed if we require congregants to enter only through a certain set of doors,” Suthar said. “While it would make temperature checks and facial recognition easier to imple-
ment, accessibility and ability to appreciate the structures get lost. Post COVID, it’s going to be a challenge across the board for any place, whether it be a house of worship or office space.”
Camera-based anomaly, suspicious object and weapons detec- tion solutions may mitigate some of these concerns by identifying threatening objects while still allowing freer use of the premises.
PEOPLE COUNTING AND OCCUPANCY MANAGEMENT
Many HoW’s still use clickers or iPads to track how many people have entered and exited their premises. This is another area where current occupancy management technology can be used to bet- ter understand traffic flows and optimize space use. Using video from strategically pointed cameras, AI analytics can provide a unified view on occupancy, use, dwell times, queue lengths and movement patterns in a designated space.
The software can also detect the same person and de-dupe the total, ensuring only unique counts are reported. These types of overlaying technologies can help a congregation control costs while still providing insights into what is happening in their physi- cal spaces.
OBJECTS LEFT BEHIND
In an ideal world, houses of worship would have the means to implement a TSA-level ability to detect objects left behind. In reality, a spectrum of options is available, from having a policy that any items left behind are discarded to again leveraging AI analytics to spot forgotten items.
Even with older image recognition technology, it’s possible to detect a delta of whether an object has been added or removed from an image via changes in field of view. Newer AI technology can take it further, with the use of object detection and the train- ing of models to recognize the most commonly left-behind items such as purses and clothing, to more suspicious objects such as bags and suitcases.
ADDITIONAL TECHNOLOGIES
LPR and smoke and fire detection are some of the other technol- ogies which can benefit from camera-based analytics. Integration of these and contextualization with other inputs (such as weath- er, etc.) can provide a comprehensive physical security solution. Whether the HoW is a small parish or a world-renown center, the ability of today’s technology to scale to a venue’s needs makes it both appealing and affordable.
While HoW’s worldwide must now face the unfortunate reali- ty that they are not immune to crime or acts of violence, there are now technology solutions which can augment traditional security practices to help manage and mitigate physical security concerns. Being able to proactively manage and monitor a space while also quickly investigating incidents can help keep these important in- stitutions open and safe to all who wish to enjoy them.
Padma Duvvuri is the Co-founder and VP of Business Develop- ment at Dragonfruit AI.
1. 2. 3.
https://www.cisa.gov/sites/default/files/publications/ Mitigating%20Attacks%20on%20Houses%20of%20Wor- ship%20Security%20Guide_508_0_0.pdf
Bledsoe, W.1966a. Man-Machine Facial Recognition: Report on Large-Scale Experiment, Technical Report PRI 22, Panoramic Research, Inc., Palo Alto, California. https://www.aclu.org/blog/privacy-technology/surveillance- technologies/amazons-face-recognition-falsely-matched-28
72
JULY/AUGUST 2021 | SECURITY TODAY
HOUSE OF WORSHIP SECURITY