International. According to Barry Norton, an executive at Milestone Systems, these contributions would be the detection and tracking of objects, the recognition or identification of objects, and the detection of anomalies.
Norton, who serves as Vice President of Research at Milestone Security, also touched on the role that Artificial Intelligence will play in future electronic security trends.
Object Detection & Tracking
Object detection is widely used in different applications, such as people counting, privacy preservation, and search based on an object's properties (such as color). There are commercial solutions based on the detection and tracking of objects, such as line crossings, perimeter protection and loitering detection.
Although object tracking is well established in scenarios ranging from simple to slightly complex, it still needs to overcome certain challenges in highly complicated scenes.
Object Recognition and Identification
In object recognition, the focus is on determining the kind of person to whom that item belongs (i.e., the class of humans), rather than considering their specific identity. However, if you want to identify each individual among the class of humans, then we are talking about identification. The maturity level of the identification technology depends on its configuration.
"In surveillance scenarios involving non-cooperative users (such as those who intentionally conceal their identity), identification becomes a challenge. However, in controlled scenarios (such as passport doors or semi-controlled areas, such as office entrances) where user collaboration is expected, the technology has reached a higher level of maturity", says Norton.
Anomaly Detection
When it comes to anomaly detection, there is debate as to whether or not it has occurred correctly. If these irregularities are well-defined (such as driving in the wrong direction or at a high speed) there are numerous commercial solutions available.
However, if the abnormalities are not clear and a generic and scalable anomaly detection system is required, which can be implemented directly in different contexts, a topic of discussion arises around whether such a product currently exists.
Trends in AI-based e-security
Understanding the scene will be a factor that will involve the analysis of supplied images to identify and segment objects and distinguish between the foreground and background of a scene.
"The main goal will be to discover the relationship between objects and take advantage of contextual information. This will allow the detection of interactions, such as those between people and between people and objects. It also facilitates the analysis of human behavior", says the Vice President of Research at Milestone Security.
In turn, the so-called multimodal analysis will be related to the evaluation of different modalities, such as conventional RGB and thermal imaging, or to the use of various sensors (e.g. cameras and access control) to analyze a given scene.
This will enable a comprehensive approach to enhance security in restricted areas, using multiple perspectives and data sources.
Finally, predictive analytics presents the most significant challenge in the field of data analytics. Its goal is to identify patterns in video data and predict future events. For example, detecting early indicators of traffic disruptions, such as a vehicle stopped in a congested lane, which has historically led to a major traffic jam.
