Sparse Reconstruction and Geo-Registration of Site Photographs for As-Built Construction Representation and Automatic Progress Data Collection

Most of the current techniques for automating progress data collection promise to eliminate laborintensive tasks associated with manual data collection. A drawback is the necessity to add additional steps to be performed before, during, or after utilization of such technologies. Working with such featureless data and without having semantic information of the scene, geometric-reasoning is problematic and induces estimation errors. In this paper application of unordered daily progress photograph logs, available on any job site, as a data collection technique is explored. In our proposed approach, a sparse 3D geometric scene of a construction site is reconstructed and photographs are geo-registered. This allows project managers to remotely explore as-built scene and geo-registered site photographs at different stages of progress, minimize their travel time, perform remote as-built analysis and use the proposed system as a tool for contractor coordination purposes. Furthermore, the point cloud allows the planned model to be registered with the asbuilt scene, in turn supporting development of the automatic 3D recognition technique and quantification of as-built progression from the geo-registered images. We present our results on two ongoing construction projects and further discuss technical issues on developing and implementing this technology for automation and visualization of as-built construction.