Video tracking

The term tracking (German for the static ( retrospective ) application equivalent to rutting, for the dynamic application ( contact following, online ) is equivalent to tracking) includes all processing steps that serve the simultaneous pursuit of (moving) objects. This is distinguished from the tracing, which relates to an offset in time tracking based on records, for example in the programming as a trace. However, the distinction is not uniform, as it is called, for example, from a GPS tracking regardless of whether the prosecution ( evaluation) is performed simultaneously or subsequently.

The aim of this persecution is usually the mapping of the observed actual movement for technical use. Such use may be the combining the tracked object to a subsequent object. Such use may also be simple the particular knowledge of the current location of the tracked object.

Tracking of an object and the modeling of the movement behavior of the surface follows the concepts of the conventional dead-reckoning navigation.

Tracking can also be a replica of a physiological or economic process concern.

Method

For tracking information about the course of the movement and the position of an object (absolute data), and on the other hand the reduction of differences ( relative error data) from a stream of observation data is extracted. Disorders arise from random technical measuring errors or unavoidable physical ( measurement noise ). Error caused by incomplete models of the replica of the actual movement, for example, assuming steady motion.

The extracted information may be, for example, the speed of movement, acceleration, and information relating to the position at a given, often lying in the future, the target point. The terms used here place, position, velocity and acceleration can be relative or absolute coordinates, so need not necessarily geographical origin. The measured variables describing the classical motion parameters of the Hamiltonian mechanics. Examples of this are continuous, such as electrical equipment for measurement data or incremental counts or discrete state information.

The quality of the specific location and motion information depends first on the quality of the observation of the tracking algorithm used and the modeling, which is used to compensate the inevitable measurement errors. Without modeling the quality of the specific location and motion information is usually disappointing bad.

The quality of the tracking is determined by the geometric and temporal resolution of the measurement equipment, the sampling and discretization and the transfer of measures from the observation. The quality of the specific location and motion information is determined by the numerical accuracy of the calculation, the iteration and integration. In addition, the determination of the integration constant of great influence.

The quality of tracking also depends upon the accuracy of the observation, ie, the measurements and the measurement error and the discretization with a finite resolution and the cyclic repetition, ie a finite sampling rate.

Intuition

First, the focus of the observation is to be sent to the relevant measure. It should be noted that the object to be observed must be detected continuously.

Deskription

Next, the current course of the observed variable is to capture and describe, for example, by a scanning.

Prediction

In this processing step, the ( computational ) prediction of the location and movement information is done based on the known history and physical or mathematical laws.

Association (also gated )

In particular, in observation rooms where usually several objects ( multi-target tracking) are and these are not uniquely identified by different measurement cycles, this component takes over the assignment of an observed in previous measurement cycles object to a currently measured object. Error in this processing step (Miss Assignments ) have a particularly heavy impact on the results.

Innovation

The determination of the current situation and other movement-relevant information is on the one hand by the prediction and the other by actual measurements (or calculations from current measurements). The innovative step leads both results are weighted together. The weighting can be both dynamic and static. A shift of the shares towards the prediction smoothes the results stronger, a greater weighting of the measurement leads to results that adapt quickly to changes in the measured values.

As a rule can be derived models that are used in model-based method for the movement patterns of the respective objects. The quality of the models or the degree of approximation to reality decisively determines the result of the tracking.

Reaction

As far as the observation should lead to an action, an appropriate response must be defined. Mostly this includes a technical adjustment of a tailored process, for example a change of a technical system behavior ( control plants) or an organizational buying behavior ( economics).

Documentation

The data collected are appropriately recorded and, where appropriate, used to improve the way forward.

Adaptation

In a sharp change in the behavior of the observed object fail traditional measuring methods, especially when the usable measuring range is left. Then the procedure of this behavior has for example by mode change, range change or be adjusted by change of pace.

Practical Implementation

In practice, tracking not always based on a one-model approach. Depending on the objects and their possible movement patterns are used for tracking an object several alternative so-called " hypotheses " set. This can be to detect and track a complicated object maneuvers, on the other hand, the weighting models can be greatly simplified by clever choice of hypotheses. The main advantage of such methods is the opposite, for example, Kalman - based methods significantly reduced computation effort. The theoretically existing larger estimation errors in phases in which the movement pattern of the objects change and leads to the " switching " of the model used is usually minimized by the parent process. Since such methods are primarily applied and further developed in the industrial and military environment, the internal details of such methods are disclosed only in part in freely accessible literature. The multi - hypotheses tracking goes back to the development of radar air surveillance systems in the 1960s.

Examples of tracking algorithms

Application examples for tracking algorithms

• Two-axis tracking photovoltaic systems
• Radar air surveillance systems ( captured by the radar flight objects are tracked in its movement. )
• Umfeldsensierung in robotics ( Covered by environment sensor objects are tracked in its movement. )
• Umfeldsensierung in the automotive sector ( Information collected by environmental sensor system objects such as cars or pedestrians are tracked in its movement. )
• Transport object detection with target of traffic flow management ( Lit.: Döring).
• Signal tracking / smoothing ( This filter smoothes measurement signals. )
• Recording of body movements ( motion tracking ) in VR applications
• Record and analyze eye movements with the eye tracker
• In acoustics: speaker tracking and fundamental frequency detection
• Attention tracking, methods for measuring attention