||As-built information of building elements (e.g. element dimension, geometry, material, etc.) could be used to facilitate multiple building assessment and management tasks, including project progress monitoring, productivity analysis, construction inspection, etc. However, the current process for retrieving as-built information of building elements from remote sensing data is labor-intensive and time-consuming. This is especially true for modeling the building indoor environments prevalent with occlusions and partitions. In order to address these limitations, the use of RGB-D mapping has been proposed and shown a promise for modeling building indoor environments. One fundamental part in the RGB-D mapping is to select an appropriate combination of visual feature detectors and descriptors. This paper investigates the effectiveness of different visual feature detectors and descriptors on modeling 3D building scenes. Several visual feature detectors and descriptors (e.g. GFTT, SURF, SIFT, ORB, and BRISK) have been evaluated. The evaluation criteria considered in the paper include accuracy and speed. The feature detectors and descriptors have been tested in multiple building scenarios with the same hardware configuration. Based on the evaluation results, it could be found that the combination of a SURF feature detector and a BRISK feature descriptor is more accurate than the others. Meanwhile, the use of the ORB feature detector and descriptor could get the fast speed.