ABSTRACT
On-site construction inspection for progress monitoring is a manual, time consuming and labour intensive process consumed by exhaustive manual extraction of data from drawings and databases. Efforts have been made to facilitate the inspection process by using emerging technologies such as Augmented Reality (AR). AR based systems can simplify and reduce the time of inspection by providing the inspector with instantaneous access to the information stored in the Building Information Modelling (BIM). However, precise alignment between the BIM model and the real world scene is still a challenge. For estimating the position and orientation of the user, methods have been proposed that either use markers or confine the user to a specific location, or use Global Positioning System (GPS) which cannot operate efficiently in an indoor environment. This paper presents an evaluation of different methods that could potentially be used for a marker-less BIM registration in AR. We implemented and tested line, edge, and contour detection algorithms using images, data from LSD and ORB Simultaneous Localisation And Mapping (SLAM) methods and 3D and positioning data from Kinect sensor and Google Project Tango. The results indicate that sparse 3D data is the input dataset that leads to the most robust results when combined with XYZ method.
