INTRUSION DETECTION AND CLASSIFICATION USING ARTIFICIAL INTELLIGENCE IN REMOTE VIDEO SURVEILLANCE

SECURITON Servis, spol. s r.o., a security management and remote surveillance company based in Slovakia has made a case study with us to test our object recognition system SAFE and evaluate the results. The goal of this study is to determine if our system can improve remote video monitoring of a facility while adding benefits such as lowering false alarms triggered by commonly used motion sensors.

THE FALSE ALARMS CHALLENGE

Despite constant advances in technology, standard security alarm systems can be a victim of their own success in that they often generate unwanted false alarms as a consequence of an inability to accurately distinguish between, for example, a stray animal, falling branch and a human intruder. In addition to the time wasted and the cost implications of having to deal with these false alarms, which might involve sending a security guard to a site to verify what may or may not be occurring, they can also be a major cause of frustration to control room operators. This has been a major issue for control rooms that have to constantly monitor hundreds of facilities, some of which generate dozens of false alarms per day.

DEKI SOLUTION

Project DEKI believes that by using the power of deep learning video analytics, we are able to provide businesses, local authorities, as well as commercially run security control rooms, with a powerful tool to help them keep one step ahead of intruders without the need of replacing existing video surveillance systems.

Most importantly, deep learning video analytics of DEKI software by default ignores video noise, waving trees, moving clouds and animals, all of which might normally be the cause of false alarms when pixel based motion detection technology or motion sensors are being used to detect activity. This ability to minimize time wasting and costly false alarms means control room operators and security personnel are able to focus on responding to real incidents and emergencies.

TESTING STATION - SECURITON

Securiton has chosen one of their clients in the city of Bratislava as a testing station for our solution. The object has 5 cameras on the outside, looking at the entrance, parking lot and the backyard of the facility. For this station, we have decided to use the DEKI SAFE hardware that can connect with up to 10 full HD standard IP cameras and give them real-time video analytics capabilities, such as detection of specific objects (person, car, etc.) and creation of virtual zones and fences. The events are locally detected and saved as metadata on the device (without the use of cloud) alongside the snapshots of the event captured by the cameras. Additionally, the notifications of intrusion have been set up to be sent via telegram to the operator's mobile device, and also connected with Securiton's existing alarm management system via remote relays.

During the course of the next months, with the help of Securiton as our testing partner, we have successfully developed and optimized person and a vehicle recognition software alongside with a virtual fencing / line crossing functionality, which has allowed us to eliminate events that are visible to the camera, but happen outside of the facility, or areas of interest.

CONCLUSION

Objects that remain in the view of the camera for multiple days (parked vehicles) can still occasionally trigger false alarms (approx. two times per night, the camera loses the object for a second due to poor light and sees it again as a new object in the area). To remove these completely, we have decided to use vehicle recognition only at the areas of entrance of the facility, where no stationary vehicles remain, but incoming and exiting vehicles still get detected. While the trade off between detection rate and incorrect alarms still exists, good light conditions, multiple viewing angles and correct setup of detection rules help ensure that all important events are recognized successfully without the need of lowering recognition sensitivity, which could create unwanted false alarms.

By the end of the testing, our DEKI SAFE successfully detected intrusion events as a simple solution of giving existing IP cameras object recognition capabilities with the use of artificial intelligence, while saving the costs of purchase and manual installation of motion sensor systems.

“From the beginning of the testing process the DEKI system was able to recognize the objects of interest (in our case persons and vehicles). The main initial problem was repeatedly recognizing stationary vehicles and in some instances not recognizing the requested objects.
The reliability of triggering an alarm in case of an object of interest entering a set area is the crucial attribute for security purposes. An environment with non standard lighting can negatively affect the recognition ability of AI. This problem was solved by adjusting the settings and by development boost which made the neural network behind the DEKI solution one of the most advanced on the market. Securiton as a testing and distributing partner has a real pleasure to work with DEKI on the projects like this because of the great communication and open minds behind the DEKI solution, but mostly because of the professional insight in the problematics of AI.”

LIBOR LAŠTÚVKA

Monitoring Center Manager

SECURITON s.r.o