In the field of Augmented Reality the distinction is made between two distinct modes of tracking, known as marker and markerless. In this blog post we are going to examine marker tracking: what it means, and what processes occur to make it possible.
A marker in the context of AR refers to what is more generally known as a fiducial marker. A fiducial marker is any object that can be placed in a scene to provide a fixed point of reference of position or scale. In AR, these markers can provide an interface between the physical world and the augmented reality content, such as 3D models or videos. At their core these markers allow the device which is generating the AR content to calculate the position and orientation (collectively known as pose), of its camera. When this calculation is done in real time, this process is known as tracking.
Marker tracking can be achieved using only visual information. In practice this means that marker tracking can be achieved with only a camera and a desktop PC; no additional sensors such as a gyroscope are necessary. In contrast, certain markerless solutions do require information from a gyroscope. For instance, our Arbitrack mode makes use of gyroscope data while our SLAM solution does not.
How is it done?
The calculation of the pose of the camera is achieved through a two-step process. First a marker is detected and confirmed. Then, using this information, the pose of the camera is calculated.
Maker detection happens in a few stages. First, to make detection easier, the camera feed is preprocessed. An important ingredient in this step is converting the camera image to greyscale. This is done for the comparative increase in the processing speed of greyscale images over colour images. In addition, the information obtained from a greyscale image of a “good marker” (see: What makes a good marker?) is perfectly sufficient for robust marker tracking.
Four fixed points are needed to determine the camera's pose. We first take the camera image data and calculate the feature points from all potential markers. We can then compare these potential markers to pre-programmed data, and check whether a potential marker matches any of the markers the device has been looking for. Once a marker has been confirmed, we use the coordinates of its feature points to calculate mathematically the pose of the camera. If you are interested in further details of this calculation, check out the link in the sources ( Section: 3.2.3 p. 51).
There are many examples of the use of markers in AR. Its prevalence over markerless tracking can largely be put down to its relative technical simplicity. It also provides a bridging opportunity which markerless doesn’t (see below). We are now going to look at two potential use-cases of markers in an app. In the first case, a marker exists only to provide a fixed point of reference; so long as it’s easily detectable, it has served its purpose. An example of this is the Karndean Designflooring iOS app, in which a marker embedded on the back of a catalogue is used in order to overlay a model of a floor.
In the second case, a marker could provide a jumping off point for the AR content. For instance, a marker could be composed of a single frame from the start of a video, but when viewed through a smartphone, the video starts playing seamlessly, and the marker image provides some context for the user of what they can expect to experience when they start to see the AR content. A great example of this can be seen below, in this app that plays a video that starts with the image from a marker.
In conclusion, markers can play an important role in an augmented reality application. When used effectively, they can provide an important bridge between your environment and AR content.
 Theory and applications of marker-based augmented reality.Espoo, VTT. 199 p. + app. 43 p. VTT Science; 3SBN 978-951-38-7449-0 (soft back ed.) 978-951-38-7450-6 (PDF) www.vtt.fi