Four functions of video analytics. Biometrics from the point of view of video analytics
Video analytics automates the four classic security features: detection, tracking, recognition and prediction. It can be argued that the order is slightly different - tracking after recognition - but in practice accurate recognition occurs after tracking an object for a period of time in order to accumulate data about an object.
As a rule, all four functions are performed repeatedly, ensuring continuous refinement of hypotheses about the number, location and type of objects in the controlled area, as well as the elimination of redundancy in the results.
Consider, for example, perimeter video analytics . On the one hand, its main task is only the primary detection of the perimeter intruder. On the other hand, a good “perimeter”, that is, video analytics for the sterile zone , performs all four functions: direct detection, tracking (to avoid repeated triggers on one object), recognition (to minimize false alarms caused by animals and other “noise” of the surrounding the world) and forecasting (for tracking in case of temporary disappearance of an object from the field).
Recognition can be understood as a wide range of tasks - from the classification of an object into a target / noise to the identification or verification of an object using biometric features.
Face recognition technology based on face biometrics is the “pinnacle” of video analytics: it sets the most difficult tasks and uses a wide range of mathematical tools. On the one hand, the biometric system implements the recognition function, establishing a probabilistic connection of the image with the identifiers of people registered in the database. On the other hand, the biometric system requires perfect operation of the other two other functions of video analytics, namely, detection and tracking.
')
Compared to other types of video analytics, the biometric system involves working with a large number of unique patterns that allow people to be identified, and imposes higher requirements on the sensor, optics, viewing angle, lighting and other parameters of the surveillance system.
Today, biometric facial recognition systems are one of the few types of video analytics that can work in crowded places. Virtually all non-biometric algorithms on the market are not able to conduct individual detection and tracking of people in a group, especially in a dense stream.
Multi- camera tracking algorithms built on biometric technology will be able to analyze the movement of people on busy objects, build their movement trajectories, estimate the residence time, and improve recognition accuracy by comparing the results from different cameras.
As a rule, all four functions are performed repeatedly, ensuring continuous refinement of hypotheses about the number, location and type of objects in the controlled area, as well as the elimination of redundancy in the results.
Consider, for example, perimeter video analytics . On the one hand, its main task is only the primary detection of the perimeter intruder. On the other hand, a good “perimeter”, that is, video analytics for the sterile zone , performs all four functions: direct detection, tracking (to avoid repeated triggers on one object), recognition (to minimize false alarms caused by animals and other “noise” of the surrounding the world) and forecasting (for tracking in case of temporary disappearance of an object from the field).
Recognition can be understood as a wide range of tasks - from the classification of an object into a target / noise to the identification or verification of an object using biometric features.
Face recognition technology based on face biometrics is the “pinnacle” of video analytics: it sets the most difficult tasks and uses a wide range of mathematical tools. On the one hand, the biometric system implements the recognition function, establishing a probabilistic connection of the image with the identifiers of people registered in the database. On the other hand, the biometric system requires perfect operation of the other two other functions of video analytics, namely, detection and tracking.
')
Compared to other types of video analytics, the biometric system involves working with a large number of unique patterns that allow people to be identified, and imposes higher requirements on the sensor, optics, viewing angle, lighting and other parameters of the surveillance system.
Today, biometric facial recognition systems are one of the few types of video analytics that can work in crowded places. Virtually all non-biometric algorithms on the market are not able to conduct individual detection and tracking of people in a group, especially in a dense stream.
Multi- camera tracking algorithms built on biometric technology will be able to analyze the movement of people on busy objects, build their movement trajectories, estimate the residence time, and improve recognition accuracy by comparing the results from different cameras.
Source: https://habr.com/ru/post/137466/
All Articles