Skip to content

Latest commit

 

History

History
64 lines (58 loc) · 4.51 KB

documentation.md

File metadata and controls

64 lines (58 loc) · 4.51 KB
Raw

Documentation

This document will give a short description of the workflow to extract tracks from recordings.

Usual Workflow

The workflow can be split in five steps:

  1. Segmentation: In this step difference images are analyzed and mosquito contours and positions are estimated
  2. Cleaning: In this step wrong detected Mosquitoes can be removed.
  3. Tracking: Tracking connects the found mosquito positions to form tracks.
  4. Cleaning: Tracks can be connected and points can be added to tracks.
  5. Exporting: The resulting position and track files can be exported. A simple visualisation is also included to save an overview image over the tracks in the observed time frame.

Parameter

Step 2. and 3. require some parameters to be adjusted.

Segmentation

MinArea: Minimum number of pixels for a blob MaxArea: Maximum number of pixels for a blob Threshold: A histogram of the difference is created and a grey scale difference threshold is calculated by taking the x percent of the distance between the value with the maximum counts to the maximum value with least a count of one.
MinThreshold: If the Threshold is below MinThreshold, Threshold will be set to minThreshold
Use the ith: Generate difference from image(n) and image(n-i)
Remove BG: Shrink the blobs by eroding with an n sized square, than dilating with the same sized stencil
Add FG: Same as Remove BG, but first dilate, than erode.
ClusterDist: This discards points too close together. Keeping only the stronger one in a radius of x
Display each N: For displaying, to speed up processing, you can skip displaying frames. Only each n-th is shown. 0 disables display completely.
Black On White: Select whether to search dark spots on light background or light spots on dark backgrounds.
medianBlur1: Apply median blur to input images before building difference image.
medianBlur2: Apply median blur to the difference image.
gaussian kernel 1: Apply Gaussian kernel to images before creating the difference image
gaussian sigma 1: "
gaussian kernel 2: Applying Gaussian kernel to the difference image
gaussian sigma 2: "
Mask Threshold: A threshold is applied to the original pictures, only pixels from 0 to threshold or (if a negative value is specified) from -threshold to 255 are selected.

Tracking

MinDist: Minimum distance travelled between frames
MaxDist: Maximum distance travelled between frames
MaxGap: Number of frames, which can be skipped while looking for the next Mosquito position
MinArea: Filter for the minimum area of the Mosquito's blob size
MaxArea: Filter for the maximum area of the Mosquito's blob size
MaxNeigh: Number of positions closest to the initial search position to analyse
MaxCohLen: Taking the maximum coherence length number of historic frames for building the prediction of the mosquitoes next position
MinTrDist:Minimum number of historical points to estimate the next position using temporal coherence
T.D.L.: Track display length, display this number long tail to the current frame in the viewer
m.T.D.L.: Minimum track display length, show only tracks with at least this number of points
append: Overwrite or append the tracking results
MaxDist(aj): For the autojoiner, set maximum distance between track end and track start for tracks to join
MaxGap(aj): For the autojoiner, set the maximum number of skipped frames

Metadata

While recording on multiple cameras, one needs to identify which frames belong together. We utilise the metadata file format from Norpix's StreamPix. To generate such files, we can use the tool txt2meta, which takes a simple text file as input, describing per line the frame number, trigger number, epoch time in seconds, milli seconds and micro seconds. As a simple example a video of 5 frames length recorded at 15 fps starting from "Mon 24 Feb 22:38:29 CET 2020" would look like:

0 1 1582580319 0 0
1 2 1582580319 66 667
2 3 1582580319 133 333
3 4 1582580319 200 0
4 5 1582580319 266 667

The usage of metadata information also allows to specify the time for frames in a video. The actual timings of the recorded video is ignored, which allows to store the video efficiently with full timing information.

Additional tools

We added a command line version of the segmentation and tracking steps, which can be found under tools directory. There is an additional tool to generate some test data, and a tool to compare position files. All tools can be build using qmake.