Data Processing of Point Clouds for Object Detection for Structural Engineering Applications

Abstract This research investigates the use of high‐resolution three‐dimensional terrestrial laser scanners as tools to capture geometric range data of complex scenes for structural engineering applications. Laser scanning technology is continuously improving, with commonly available scanners now able to capture over 1,000,000 points per second with an accuracy of ∼0.1 mm. This research focuses on developing the foundation toward the use of laser scanning to structural engineering applications, including structural health monitoring, collapse assessment, and post‐hazard response assessment. One of the keys to this work is to establish a process for extracting important information from raw laser‐scanned data sets such as the location, orientation, and size of objects in a scene, and location of damaged regions on a structure. A methodology for processing range data to identify objects in the scene is presented. Previous work in this area has created an initial foundation of basic data processing steps. Existing algorithms, including sharp feature detection and segmentation are implemented and extended in this work. Additional steps to remove extraneous and outlying points are added. Object detection based on a predefined library is developed allowing generic description of objects. The algorithms are demonstrated on synthetic scenes as well as validated on range data collected from an experimental test specimen and a collapsed bridge. The accuracy of the object detection is presented, demonstrating the applicability of the methodology. These additional steps and modifications to existing algorithms are presented to advance the performance of data processing on laser scan range data sets for future application in structural engineering applications such as robust determination of damage location and finite element modeling.