A security systems provider was looking to improve the accuracy of their event detection technology. Their existing technology was able to detect when their parameters were breached but it wasn’t able to sense whether it was a human breach or simply environmental, such as leaves, branches, animals, or weather conditions. They needed to find something that worked with their Distributed Acoustic Sensing technology (DAS) that could monitor activities across kilometers of length that used fiber sensors to gather information.
We decoded the historical data stored on the company’s sensors using Deep Learning algorithms to help understand their breaches. This helped us understand a range of intrusions based on several circumstances in different conditions. As a result, our technology was able to monitor a breach within 5 seconds and accurately trigger an alarm 98% of the time. This meant that the client could spend less time and resources investigating false alerts. Due to the heightened level of accuracy, they were also able to use the technology in harsh or active environments without worrying about false breaches.
Under a controlled context, we used different data-related methods to better understand historical files on their existing systems to label each intrusion logged. Due to the nature of the data, we also applied data augmentation techniques to further train algorithms using a deep learning approach that would improve surveillance. Once all of the data from testing was gathered, we were then able to test it in a real-life context and environment.
By the completion of this project, we created a system that was able to reduce the number of false positives significantly compared to the client’s pre-existing system. It was able to accurately detect real intrusions as opposed to environmental events. This lead to higher tracking accuracy and helped the security team save time and resources on investigating false positives.