Detection and tagging of terrain features in camera images
For several years, so-called geo information services enjoy increasing popularity. To some extent this was made possible by the wide availability of powerful computing technology and broadband internet connections. This combination provides the user with (in many cases free) access to large databases of location based information and means of displaying this data in an appealing way. In parallel to the advancements in software technology, a tremendous fall in the price of GPS receivers allowed for all new applications, ranging from navigation devices to modern smartphones providing access to location aware services.
Particularly regarding the linking of current location information with other data, augmented reality is an important catch phrase. It describes the superimposition of virtual objects on camera images in such a way that those objects seamlessly integrate with the image of the real environment. Lately the term is often misused for products that are only capable of overlaying location based information without referencing them to the actual image.
In the course of this thesis a technique for tagging terrain features in camera images has been developed. It achieves the referencing of the virtual data to the image contents by combining signals from several sensors and the results of image processing. In particular, the main objective this work pursues is to mark and to designate points in images of mountainous areas that are relevant to tourists, hikers, and mountaineers.
For this purpose the alignment of the camera in space is computed from data acquired by a magnetic field sensor and an acceleration sensor. The camera attitude, in conjunction with the position determined by a GPS receiver, allows for the computation of a virtual image of the observed terrain based on a digital elevation model. As this alignment measurement is not accurate enough for the correct positioning of virtual elements in the camera image, it gets adjusted by means of image processing. To accomplish this, the horizon line, as a distinctive feature, is detected in the camera image and then correlated with its virtual equivalent.
This method provides a transformation matrix that allows for the conversion of geographic coordinates to positions in the camera image. Therefore the possible applications with respect to the type of information to be displayed are almost limitless.
Keywords: augmented reality, digital elevation model, DEM, terrain feature, camera image, GPS, horizon line
September 21, 2012