Set Up Tracking

Set Up Tracking#

General Settings#

At the top of the tracking tab, you can find the general settings of tracking. In the end you have two options to apply the tracking to your project.

online calculation

Enables the tracking, but you manually have to navigate though the video.

calculate all

Enables the tracking and runs the entire video forward and backward automatically

However, other settings should be fixed beforehand.

repeat below quality

If the pedestrian already has a tracked trajectory (e.g. when running the video backwards, after running it forwards), tracking is only repeated if this is checked and the quality is below the threshold. Otherwise, the existing point is not overwritten.

extrapolation for big diff.

If there is a huge difference in speed from one frame to another, discard the value and instead extrapolate. Note that extrapolation is at most applied 3 times in a row. After this, the trajectory is terminated.

merge

Allows you to merge two trajectories into one continuous trajectory. See more.

only selected

Enables that tracking settings will only be applied to the selected subjects. This works only in combination with the show only people or the show only  people list option in the path section at the bottom of the tracking tab.

export

Exports trajectories in desired format.

import

Imports trajectories from files.

reset

Deletes all trajectories. This is useful during tuning of parameters to allow a view of the current settings (recognition and tracking) applied to the current frame.

Region Of Interest#

The region of interest (ROI) is the part of the image where the tracking is performed. For the ROI we see the options

show

Shows the ROI as a blue rectangle on the image.

fix

“Fixes” the rectangle, such that it cannot be moved with the mouse.

full image

Sets the ROI to the full image.

adjust automatically

Sets the ROI as one head size larger than the recognition ROI

Warning

The tracking ROI must be larger than the recognition ROI. Any detection outside the tracking ROI will create an entirely new, one point trajectory.

Export Settings#

In this section, various export settings can be chosen. For more information about the settings as well as information on importing and exporting files, check out the Trajectory Import and Export section.

Search Region#

scale

Adapts the window size of the search region in which the Lucas-Kanade[1] method is applied.

levels

Adapts the number of search regions.

max. error

Adapts the error acceptance value. A person is no longer tracked if the calculated error of the tracking is higher than an acceptance value.

show pyramidal search size

Enables the visual representation of the search regions.

adaptive level

Enables the re-tracking at a lower pyramid level for failed tracking attempts. Mostly leads to poorer trajectories!

Tip

If you are tracking pedestrians with colored hats, make sure to include textured areas (e.g. the border of the hats) into the search region. There the hat displays different intensity levels of color which are essential to the Lukas-Kanade method.

After you have set all of these settings, you can run the tracking with either online calculation or calculate all as mentioned on top of this page.

Important

You need to manually export the trajectories that resulted from tracking. You should export in the .trc-format, since it is the working file format of PeTrack.

Path Settings#

In this section, the depiction of the trajectories as well as additional information about them can be adjusted.

show only people

Only the trajectory of the selected person will be displayed. The number represents the PeTrack ID which are assigned in an ascending order.

show only people list

Only the trajectories of selected people will be displayed. You can either select the PeTrack IDs from the list or write them in the box separated by commas. Ranges are also possible, i.e. 7,15-20 is a valid entry and shows person Nr 7 and persons from 15 to 20 (including ends).

frames before/after actual position

Number of frames before/after the actual position for which the trajectories should be visible.