Feature Detection and Tracking

Much of the data extraction and processing performed by 2d3 technologies depends on the identification and tracking of two dimensional features from a source image set. The source can be a static image set or a moving image sequence.

2d3's automated feature detection and tracking algorithms identify hundreds or even thousands of features in a scene and follow their motion throughout a sequence. We have tools which allow for adjustment of the feature identification process including its general sensitivity, the image area over which searching is performed, the colour channels within which the features are identified and the size of identified features. The size of the search area used can be adaptive which means that the search area increases automatically if a feature cannot be found in contiguous frames.

We also have tools for user identification of 2D features and the ability to manually track them.

Our standard automated method of feature identification uses a form of corner detection but we also have the capacity to pattern match and track shapes as they change in two dimensions with their relative orientation to a moving camera. This is a process which can be useful in green or blue screen chroma key or similar scenarios where markers are placed in the environment to identify features in the scene.

In addition we have technology which allows the automatic identification and reading of fiducial style markers containing information which can be used for camera calibration and tracking in real-time or across a number of still images with wide spatial separation.