Tracking

Tracking (German for the static [retrospective] application synonymous with track formation , for the dynamic application [running, online] synonymous with tracking ) includes all processing steps that serve to track (moving) objects at the same time. A distinction is made between this and tracing , which concerns a time-staggered tracking based on recordings, e.g. B. in programming as tracing . The delimitation is not uniform, so one speaks z. B. from a GPS tracking regardless of whether the tracking (evaluation) takes place at the same time or afterwards.

The aim of this tracking is usually to map the actual movement observed for technical use. Such use can be the merging of the tracked object with a subsequent object. Such use can, however, also be more simply knowing the current location of the tracked object.

The tracking of an object and the modeling of its movement behavior on the surface of the earth follows the concepts of classic dead reckoning .

Tracking can also relate to the simulation of a physiological or economic process .

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 deviations (relative error data ) are extracted from a stream of observation data. Interferences arise from random technical measurement errors or from unavoidable physical (measurement noise). Errors arise from incomplete models of the simulation of the actual movement, for example assuming stationary movement.

The extracted information can be, for example, the speed of the movement, the acceleration and information relating to the position of a specific target point, which is often in the future. The terms place , position , speed and acceleration used here can be relative or absolute coordinates, so they do not necessarily have to be of geographical origin. The descriptive measured quantities are the classic motion parameters of Hamiltonian mechanics . Examples of this are continuous, such as electrical, measurement data or incremental counter values or discrete status information.

The quality of the determined position and movement information depends first of all on the quality of the observation, on the tracking algorithm used and on the formation of the model, which serves to compensate for unavoidable measurement errors. Without modeling, the quality of the specific position and movement information is usually disappointing.

The quality of the tracking is determined by the geometric and temporal resolution of the measuring equipment, the scanning and discretization and the transfer of the measured variables from the observation. The quality of the determined position and movement information is determined by the numerical accuracy of the calculation, its iteration and integration. In addition, the determination of the integration constants is of great influence.

The quality of the tracking also depends on the accuracy of the observation, i.e. the measurements or the measurement errors as well as the discretization with a finite resolution and that of the cyclical repetition, i.e. a finite sampling rate .

intuition

First of all, the focus of the observation is to be directed to the relevant measured variable. It must be taken into account that the object to be observed must be recognized continuously.

Description

The current course of the variable to be observed is also to be recorded and described, for example by a scanning function.

Prediction

In this processing step, the (computational) prediction of the position and movement information takes place on the basis of the known history and physical or mathematical regularities.

association

In particular in observation rooms, in which there are usually several objects (multi-target tracking) and these cannot be clearly identified via different measurement cycles, this component assumes the assignment of an object observed in previous measurement cycles to a current measurement. In order not to have to check all measurements with an object for a possible association, remote measurements can be excluded from the outset using heuristics. This step is called gating and is part of the association step.

Errors in the association step (so-called missassignments or incorrect assignments) have a particularly severe effect on the results.

innovation

The current position and other information relevant to movement are determined on the one hand by the prediction and on the other hand by current measurements (or calculations from current measurements). The innovation step brings both results together in a weighted manner. The weighting can be done dynamically as well as statically. A shift in the proportions towards the prediction smooths the results more strongly; a greater weighting of the measurement leads to results that adapt more quickly to changes in the measured values .

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

reaction

If the observation should lead to an effect, a corresponding reaction must be defined. In most cases, this also includes a technical setting of a tailored process, for example a change in technical system behavior (control systems) or organizational purchasing behavior (economy).

documentation

The recorded data are expediently recorded and, if necessary, used to improve the further procedure.

Adaptation

In the event of sudden changes in the behavior of the observed object, classic measuring methods fail, especially if the usable measuring range is exceeded. Then the procedure has to be adapted to this behavior, for example by changing modes, changing areas or changing cycle times.

Practical implementation

In practice, tracking is not always based on a one-model approach. Depending on the objects and their possible courses of movement, several alternative so-called "hypotheses" are used to track an object. In this way, on the one hand, complicated object maneuvers can be recorded and tracked, and on the other hand, the weighting models can be greatly simplified if the hypotheses are skillfully selected. The main advantage of such methods is that compared to z. B. Kalman-based method significantly reduced computational effort. The theoretically larger estimation error in phases in which the movement of the objects changes and leads to the "switching" of the model used is usually minimized by higher-level processes. Since such processes are primarily used and further developed in the industrial and military environment, the internal details of such processes are only partially disclosed in freely accessible literature. Multi-hypothesis tracking dates back to the development of radar aerial surveillance systems in the 1960s.

Examples of tracking algorithms

α / β / γ filter: A model-based method that estimates the movement parameters of the observation object using a simplified model
Kalman filter: A model-based method that estimates the movement parameters of the observation object
Sequential Monte Carlo method (particle filter): is suitable for tracking if the system has non-Gaussian, non-linear dynamics.

Application examples for tracking algorithms

Two-axis tracking of photovoltaic systems

Radar aerial surveillance systems (objects in flight detected by the radar are tracked in their movement.)

Environment sensing in robotics (objects detected by environment sensors are tracked in their movement.)

Environment sensing in the automotive sector (objects detected by environment sensors such as cars or pedestrians are tracked in their movement.)

Traffic object detection with the aim of traffic flow control ( Lit .: Döring).

Signal tracking / smoothing (The filter smooths measurement signals.)

Detection 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

AttentionTracking , method for measuring attention

Single particle tracking

literature

Samuel S. Blackman, Robert Popoli: Design and Analysis of Modern Tracking Systems (Artech House Radar Library) . Artech House Inc, London 1999, ISBN 978-1-58053-006-4 .

Individual evidence

↑ Movement parameters of the observation object (PDF; 110 kB) on Buffalo.edu

↑ Kalman filter (PDF; 178 kB) on UNC.edu

[1] Movement parameters of the observation object (PDF; 110 kB) on Buffalo.edu

[2] Kalman filter (PDF; 178 kB) on UNC.edu