False alarm detection has become a critical function in modern surveillance monitoring systems.
Cloud or onsite
With all the talk around cloud systems and solutions which will apparently solve all your problems no matter what industry you are in, we first asked Alkema which solution he thinks is best. He says, “Effective monitoring of a site is a critical aspect of any security system, and both cloud and onsite systems have their advantages and disadvantages when it comes to reducing false alarms.”
“Cloud-based systems are becoming increasingly popular due to their flexibility and scalability. However, cloud-based systems are dependent on Internet connectivity and may not be as reliable as onsite systems in areas with poor connectivity. Cybersecurity considerations must also be assessed carefully to ensure that the customer’s risk is managed when deploying a cloud-based solution.”
Alkema adds that onsite systems could be viewed as more reliable than cloud-based systems since they are not dependent on an Internet connection if the site is monitored on the site itself. However, it’s worth noting that the once-off and maintenance costs of onsite systems could increase more than cloud-based systems depending on their cost base. “It is critical to assess each customer’s requirements and their connectivity restrictions individually before deciding on solutions.”
On the question of bandwidth, he notes, “Bandwidth requirement is directly impacted on how the solution is designed. You could deploy edge-based, on-camera AI, where the camera only sends events and video once it has determined there is something to report. This could save a lot of bandwidth and computing resources, whether it is connected to a cloud or onsite based solution. It comes back to how the customers want to monitor their sites, the requirements of the customer, and connectivity constraints they need to consider when designing a solution.”
You have to train people and machines
We hear a lot about false alarm detection applications reducing false alarms by 90% or more (although few people seem to focus on how many real events systems miss), however, the efficacy of any AI is dependent on the training it receives – similar to people. It’s easy enough (these days) to distinguish a plant blowing in the wind from a human, but how do the systems differentiate between, for example, people hugging vs. people fighting?
Alkema agrees that in-depth training of video-based AI is essential but adds that the data used to train the AI will also directly impact the outcome of how it will perform. He says vendors balance the following aspects when training neural networks:
- Data quality: The quality of the data used to train the neural network is critical. Poor quality data or badly tagged data can lead to inaccurate results and can negatively impact the performance of the neural network.
- Data quantity: Depending on how specific you are in training your neural network, large amounts of varied data would be required to train neural networks effectively, which can be time-consuming and expensive.
- Computational resources: Training video-based neural networks requires high-end costly computational resources.
Losing jobs
The question of how much we can rely on AI to do the jobs that required trained operators in the past is also a question that needs to be addressed. The question of how many jobs AI will take over from humans is quite a popular discussion point in all industries as it has become evident that it won’t only be blue-collar workers in the unemployment line.
Looking specifically at human control room operators, how ‘endangered’ would AI make their jobs? Currently the best scenario is to have a black-screen approach where AI removes most of the false alarms and hopefully catches all the real alarms and notifies an operator, who then implements the standard operating procedures (SOPs) for the company or for that specific client (if the control room monitors multiple clients’ infrastructures).
Currently AI can’t take over those jobs, or can it? In the future, however, what is to stop an AI from recognising a real event and implementing those SOPs automatically without having to rely on humans to finish the job?
“Even as AI technology matures,” says Alkema, “human operators will still be needed to verify and act on events generated by AI. Some actions could be automated by video management systems without human intervention, but this would depend on the customer’s policy in responding to actual events. The ideal risk mitigation strategy would incorporate the best of what AI-based automation and human intervention can offer, so that operators can use these tools to be even more effective.”
However, he also notes, “There is so much investment in video-based AI on a global level. The rate at which AI technology is improving, while becoming more accurate and efficient, will change the landscape of what a monitoring control room will look like in the future.”
CathexisVision’s AI and false alarm options
Finally, we asked Alkema what Cathexis offers in the realm of false alarm detection.
“At Cathexis, our ethos in developing AI-based solutions is to enhance the value of video data. A key way in which we do this is with the CathexisVision video analytics suite, which uses neural network models and AI to ensure that the software can detect anomalies onsite, and only alert control room operators to events they need to be informed about. Additionally, there is CathexisVision’s powerful event handling tool, the Alarm Management Gateway, which allows operators to view both live and recorded video events, be informed of their SOPs, and report on the actions they have taken.”
“Reporting on event handling is an important feature in the software, which helps to maintain control room integrity. CathexisVision’s effective implementation in major, high-traffic sites globally is a testament to the solution’s ability to enhance the value of video data, so that operators can stay informed and focus on the most essential tasks.”