Application of segmentation algorithms in terrestrial laser scanning to enhance the accuracy of area calculations of architectural structures

Abstract

The article explores the problem of increasing the accuracy of area calculations for architectural objects based on Terrestrial Laser Scanning (TLS) data. The aim of this study is to develop a point cloud processing method that minimizes metric errors without constructing a full-scale 3D model. Specifically, it seeks to improve area measurement accuracy by complementing the manufacturer-provided laser scanning data processing algorithm with an additional workflow utilizing standard tools to minimize and control measurement errors. This approach is demonstrated using scanning data from a residential building restored after shelling. Such objects are particularly challenging for measurement due to atypical deformations, structural damage, wall deviations from the vertical, and the complex geometry of residual destruction. Under these conditions, the application of standard, fully automated data processing templates is highly inefficient and frequently leads to erroneous results. Therefore, to reduce data processing time, the key element is preparing the point cloud for geometric parameter measurement rather than generating a 3D model. The methodological basis of the work consists of general scientific and specialized research methods, in particular, formal-logical, systemic-structural methods, and the method of analysis and generalization of scientific sources and forensic practice. The research was conducted using scanning data from a residential building restored after being damaged by a drone. The proposed method is based on combining the Region Grow segmentation algorithm within the Leica Cyclone 3DR environment with statistical analysis of deviation histograms to verify the constructed approximating plane. A key element of the approach is the isolation of a “trust zone” of points belonging to the target plane, instead of a subjective selection by the operator. The scientific novelty lies in a comprehensive approach to the study of the scanned point cloud as an independent object of forensic activity, identifying key problems in their research, and evaluating the results obtained. The conclusions emphasize that applying the proposed method allows for obtaining reliable results with an error of ±0.002 m, which corresponds to the LOA 40 (Level of Accuracy) according to the international USIBD (U.S. Institute of Building Documentation) standard. The practical value of the work lies in the possibility of applying the results as a legally verified evidentiary base in forensic expertise, as well as for calculating restoration estimates and damage assessments. The proposed approach significantly reduces processing time compared to the standard Scan-to-BIM process and ensures the objectivity of measurements on complex architectural objects.