Employing Deep Learning to Enhance Surveillance

By CIOReview | Thursday, March 1, 2018
214
345
61

From AI and deep learning to advances in privacy protection, extensive video-surveillance innovations are on the rise around the world. Over the last couple of years, there has been a tremendous surge in research and advances surrounding deep learning. Deep learning algorithms are employed into fully deployable products, with user-friendly interfaces and scenario-focused solutions. There are two main areas where deep learning analytics offer great benefits: accuracy and power. Deep learning algorithms have the ability to view a scene intuitively, increasing accuracy, and reducing false alarm rates. The algorithm makes it possible to process and analyze vast amounts of video footages into usable information in a short amount of time. There are video processing software applications that use deep learning algorithms, allowing users to interact with surveillance footage using a Google-like interface and natural language search terms. This drastically reduces the time taken to find relevant video footage from thousands of feeds. Facial recognition is another area that has benefited immensely from deep learning. For instance, deep learning face recognition algorithms are embedded in search engines to find missing people from video footages.

IC Realtime, a designer and manufacturer of Advanced Video Surveillance systems and software applications, created ‘Ella,’ a new cloud-based search engine that uses deep learning algorithms to enhance surveillance systems with natural language search capabilities across recorded video footage. Ella provides surveillance and security camera systems the ability to recognize objects, colors, people, and vehicles instantly, enabling users to find relevant clips in seconds. Over the next few years, deep learning algorithms will enable video analytics applications to run on servers. The future of video analytics will also see the incorporation of AI in enabling multiple systems to communicate with one another to take decisions, quickly identify suspicious activities, or predict them before they happen, and in creating an event and authorization-based alert systems.