Automated Cleaning of Point Clouds for Highway Retaining Wall Condition Assessment Pedram Oskouie 1 , Burcin Becerik-Gerber 2 and Lucio Soibelman 3 1 PhD Student, Sonny Astani Department of Civil and Environmental Engineering, University of Southern California, Los Angeles, CA; PH (213) 572-9373; Fax (213) 744-1426; email: oskouie@usc.edu 2 Assistant Professor, Sonny Astani Department of Civil and Environmental Engineering, University of Southern California, Los Angeles, CA; PH (213) 740- 4383; Fax (213) 744-1426; email: becerik@usc.edu 3 Professor and Chair, Sonny Astani Department of Civil and Environmental Engineering, University of Southern California, Los Angeles, CA; PH (213) 740- 0609; Fax (213) 744-1426; email: soibelman@usc.edu ABSTRACT Continuous condition monitoring and inspection of under-construction highway retaining walls is essential to ensure that construction performance criteria are met. The use of LIDAR systems by the construction industry has been significantly increased in recent years especially for recording the as-built and as-is conditions of facilities. The high-precision, dense 3-D point clouds generated by 3-D laser scanners can facilitate the process of Asset Condition Assessment (ACA). ACA involves preprocessing the point cloud data, for which point clouds have to be cleaned of any unwanted or occluding objects, and noises. As part of this research, the retaining wall point cloud data from an ongoing highway construction project was processed and analyzed. The project uses 3-D laser scanners for regular monitoring of mechanically stabilized earth walls that retain the soil supporting the highway alignment. The temporary steel and wooden brackets that hold formworks on top of the walls along with other construction materials are defined as unwanted objects. The authors have used a non-deterministic algorithm to remove the brackets and noises from the point clouds. Various settings of the algorithm have been analyzed using different sets of data. This paper presents the accuracy and performance of the tested algorithm and its evaluation when comparing the results with manually cleaned point clouds. INTRODUCTION With the advent of remote sensing technologies, 3-D laser scanners have been employed by the construction industry for several different uses such as 3-D as-built model reconstruction (Tang et al. 2010, Xiong et al., 2013), project control and progress monitoring (Turkan et al. 2012), MEP clash detection and as-built reconstruction (Tang et al. 2013, Bosche et al. 2013), 3-D thermal modeling (Wang et al. 2012), and infrastructure surveying and inspection (Tang et al. 2012, Olsen et al. 2012). Laser scanners enable rapid and accurate data collection from under- 966 COMPUTING IN CIVIL AND BUILDING ENGINEERING ©ASCE 2014