visiona is an optical coded target tracking library. It allows to precisely and accurately determine the relative position and orientation of a target with respect to a calibrated camera.

Some of the principles are described in details in the following (open-access) publication, along with test results (download link):

@article{cucci2016accurate,
  AUTHOR = {Cucci, D. A.},
  TITLE = {ACCURATE OPTICAL TARGET POSE DETERMINATION FOR APPLICATIONS IN AERIAL PHOTOGRAMMETRY},
  JOURNAL = {ISPRS Annals of Photogrammetry, Remote Sensing and Spatial Information Sciences},
  VOLUME = {III-3},
  YEAR = {2016},
  PAGES = {257--262},
}

This software was developed within the scope of the mapKITE project, funded by the European Commission through the Horizon 2020 program. The copyright holder is the École polytechnique fédérale de Lausanne (EPFL).

An example of the coded targets employed by visiona is reported below.

The position of the white dots in the black annulus encode a signal that is used for robust target detection.

visiona currently relies on configuration files to specify target properties (such as the dimension and the code) and the intrinsic camera calibration.

A target example, ready to be printed and employed, can be generated with the target/target.tex LaTeX file. A configuration file which would allow to detect such a target is given in config/sample.cfg.

TODO: add a target/config generation script.

One should first include the main visiona headers:

#include "Visiona.h"

To detect a target in an image, we first need to read the configuration file. Configuration file parsing is done via libconfig. Here is an example

MarkerDetectorConfig cfg;
cfg.loadConfig("config/sample.cfg")

Then a MarkerDetector object has to be created

MarkerDetector *detector = MarkerDetectorFactory::makeMarkerDetector(cfg);

Target detection is achieved by

vector<shared_ptr<Target>> targets = detector->detect(raw);
shared_ptr<Target> target = ret[0];

where raw is an opencv Mat object. The current implementation searches for a specific target whose code is known a-priori. The library is ready to detect and measure multiple different targets in the same image, but minor modifications to the code are required.

The returned Target objects accumulate the results of the detection and measurement steps.

Optional: black/white levels can be accurately identified with. This improve pretty much everything that comes later. If this method is not called before measurements, 0 (black) vs 255 (black) levels are assumed, which is never the case.

detector->evaluateExposure(raw, target);

Two main methods are available for target measurement. The first is based on the solution of the PnP problem using the code points. It is fast and suited for real-real time applications:

detector->measureRough(raw, target);

translation = target.rought; // cv::Mat
orientation = target.roughR; // cv::Mat

A more precise measurement of the target center and distance can be achieved with:

detector->measure(raw, target);

distance = target.distance;
cv::Point2f center(target.cx, target.cy);

An application is included in the current source tree, called batchdetector, which allows to traverse a directory and process all the contained images. Refer to this application for degub options, visualization, timing statistics and other details.

visiona depends on opencv2, libconfig, eigen3.

The source code is released under a GPLv3 licence.