Summary - Experiments with the Layar augmented reality browser by showing a 3d image of a building internal structure superimposed over the camera view. In this evaluation we set up a service that could display 3d images in the Layar augmented reality browser [6]. The model adopted by the Layar framework is that the Augmented Reality browser application is provided by Layar “out the box” for users to download to their iPhone or Android phone. Users can then browse “Layars”, which are published collections of “points of interest” which the Layar browser is able to access from the web and display as text or images on the camera view. The “point of interest” content provider must make their data available following the Layar API specification for the browser to consume using their own hosting software and infrastructure. The content provider must also publish metadata to the Layar site so that the user can discover (e.g. search for ) Layars using the mobile client.
We were fortunate to have some help from an architect called Chris Lowry from the Edinburgh College of Art who has a teaching method called "Building Anatomy" where 3d models are used to show students the intricate details of the inner structures behind building facades. We tried implementing one of Chris models’ in a Layar provider service so that the 3d model could be superimposed on the camera view as the student holds up the device in front of the actual building.
The first problem to overcome was adapting the rich 3d model from Chris Lowry’s study into a format that the Layar browser could consume more easily. After a fair bit of manipulation to the 3d model we did get something working and learnt some interesting lessons around the process. First we established that some degree of technical knowledge and infrastructure is necessary to publish to Layar. At a minimum, the content provider needs to implement the Layar API as a web service using a server side language such Java. The documentation for the API was good, but not always 100% accurate – for example some fields specified as mandatory in the specification were not in fact included in the Layar browser request and it was clear from frequent updates to the wiki that the API specification was still in a fair amount of flux.
A more pressing issue for an application such as BuildingAnatomy is the accuracy of GPS in urban landscapes. In built up areas GPS accuracy was less than 70m whereas we estimate we need 0.5-1m accuracy to align a 3d model to a building façade. Our workaround was to enter the location manually using the Layar Settings page which is fine for testing but not a compelling solution in the field.
We have a few ideas about ways to overcome this. One involves setting up predefined vantage points (perhaps using QR codes [12]) to obtain an accurate GPS reading that could be used for viewing a point of interest from a known location or to calibrate the device. Another approach that has been adopted elsewhere [13] is to employ 3d image recognition techniques to accurately pinpoint the users location relative to known building footprints. We recommend further investigations into these techniques as clearly improve the accuracy of the device geo fix is crucial to many teaching and learning applications of AR.