Professional Analytics has several types of Object Trackers. The Object Tracker is the core component of the analytics solution and is the one that recognizes the objects in the scene, determines their speed, size, direction... Each Object Tracker has different techniques for analyzing a scene and identifying objects. Once objects are identified and classified, rules can be applied to these objects in order to trigger alarms. Each type of tracker is capable of recognizing different types of objects and features.
The configuration of which tracker to use is in Advanced Options of the Analytics Configuration:
•Object Tracker: This is the standard object tracker. This tracker uses only CPU (Does not use GPU) and uses the movement to differentiate the background and foreground objects in the scene, together with the calibration grid and object classification options to recognize and classify objects. As only motion is used to detect an object, the tracker itself is not recognizing an object by the way it looks, instead an object's classification is defined through the use of its estimated size and speed (which are determined by correct calibration of the 3d grid).
•Object Tracker (Deep Learning): This object tracker utilizes appearance based neural networks, trained on millions of images, to locate and recognize similar objects in a scene. As a result, the motion of a tracked object and its estimated speed and size are not required to determine the object's classification. The use of appearance based deep learning models generally provides more accurate and diverse object detection in a scene and tends to have lower levels of false alarms due to their resiliency to environmental effects such as lighting changes or extreme weather. This tracker is the most suitable tracker for general analytics use cases if you have a compatible GPU on the analytics server, however it is not recommended for use with thermal cameras, where the standard Object Tracker is the most recommended for this case. Due to the nature of deep learning and specifically neural networks this type or processing is done on GPU.
•People Tracker (Deep Learning): This tracker is suitable for processing scenes with people, and just like the Deep Learning Object Tracker, it also uses pre-trained neural networks to recognize people. In this case the People Tracker's Deep Learning models are specifically trained and optimized to only detect people and help identify human behavior.
•Skeleton Tracker (Deep Learning): This tracker is suitable for tracking people in situations where the camera's field of view is relatively close. The Skeleton Tracker is based on Pose Estimation technology, providing the location of the person in the camera's field of view, as well as additional data such as body parts that are used for advanced pose recognition rules.
•Hand Object Interaction Tracker: The Hand Object Interaction tracker was developed to detect hands and the objects they hold. This tracker requires a top-down, relatively close field of view to detect optimally. This tracker requires an additional Behavior license.
Some rules are only available to a certain type of tracker, such as the Fall rule (Fallen People) can only be used with the People Tracker (Deep Learning) and Skeleton Tracker (Deep Learning). Select the ideal tracker for your type of scenario.