Change detection for indoor construction progress monitoring based on BIM, point clouds and uncertainties

Highlights

• A new method for fusion of model uncertainty with uncertainties of point clouds.

• Consideration of indoor scanning geometry for accuracy assessment.

• Voxel based change detection and self-occlusion analysis.

• Experiments on progress evaluation on two real construction sites.

Abstract

Automatic construction progress documentation and metric evaluation of execution work in confined building interiors requires particularly reliable geometric evaluation and interpretation of statistically uncertain as-built point clouds. This paper presents a method for high-resolution change detection based on dense 3D point clouds from terrestrial laser scanning (TLS) and the discretization of space by voxels. In order to evaluate the metric accuracy of a BIM according to the Level of Accuracy (LOA) specification, the effects of laser range measurements on the occupancy of space are modeled with belief functions and evaluated using Dempster and Shafer's theory of evidence. The application is demonstrated on the point cloud data of multi temporal scanning campaigns of real indoor reconstructions. The results show that TLS point clouds are suitable to verify a given BIM up to LOA 40 if special attention is paid to the scanning geometry during the acquisition. The proposed method can be used to document construction progress, verify and even update the LOA status of a given BIM, confirming valid and BIM-compliant as-built models for further planning.

Introduction

Change detection in terms of construction progress documentation and evaluation is currently of great interest for the AEC (Architecture, Engineering and Construction) branch and BIM (Building Information Modeling). Not only many positive experiences in construction practice indicate a strong correlation between permanent construction monitoring and compliance with project schedule, but careful researches and case studies such as presented in [2] confirm this positive effects also from a scientific point of view. Nevertheless, progress monitoring and evaluation during the construction phase are usually time-consuming and only poorly automated.

So far progress monitoring mostly comes with visual inspections performed by construction workers or time and cost-intensive professional surveys using total stations and leveling devices. 3D laser scanning seems a good alternative for fast and high resolution ”as-built” recording on site and there are many approaches and techniques for reconstruction from point clouds, e.g. assembled and discussed in [43]. However, this technique still lacks deliberate 3D point cloud analysis, computer aided interpretation and strategies for the evaluation of metric tolerances in engineering practice. In a building constructions interior, monitoring effort based on 3D point clouds is even greater and therefore currently still unprofitable. This is due to very limited views in constraint indoor environments, which usually cause weak scanning geometries and occlusions. At the same time there is a great demand for evaluating detailed interiors which require even higher geometric resolution and reliably confirmed accuracy. In this paper, we address this challenge and introduce a method for automated change detection based on laser range measurements, which enable assumptions about the occupancy of indoor space (Fig. 1).

The remainder of this paper is organized as follows: Section 1 discusses related literature and presents the main contributions of the work, including a method overview. Section 2 introduces fundamental topics concerning the overall approach before moving on to the basic concept of voxel-based change detection in Section 3. The main part of the paper is represented by Section 4, which explains the point cloud-to-BIM verification method, based on TLS measurements and evidental reasoning with statistically uncertain reference geometry from a given BIM. Based on the TLS measurements of two real indoor conversion projects, the practical applicability of the method is demonstrated by experiments in Section 5, followed by a discussion of the results in Section 6. Section 7 gives a final summary, concludes the paper and presents plans for future research.

Change detection: Dense 3D point clouds can be analyzed to derive knowledge about the occupancy of space through voxel-based discretization. The origins of change detection based on 3D point clouds and occupancy grids are mainly in the field of informatics and robot navigation and were first introduced by [28]. Due to the practicality of this method, it has also already been used for several years in the field of engineering and geodesy to process high-quality point clouds from multi view photogrammetry and 3D laser scanning. [1] process 3D point clouds based on occupancy grids for change classification in urban landscapes whereas [22] apply near-real-time detection for monitoring landslides from laser scanning point clouds. Also related to large-scale areas is the work of [15], who presented a framework for automatic change detection from airborne laser scanning (ALS) point clouds. They used a relatively coarse voxel space for comparing point clouds from different epochs and refined the change detection result by considering single points in pre-selected areas of interest. The impact of a laser range measurement was modeled by belief functions according to Dempster Shafer theory [8,34], which reflected the uncertainties and physical properties of a typical ALS measurement. [11] build up on the work of [15] and combined change detection based on single points and occupancy grids for mobile laser scanning data in order to efficiently handle occlusions and at the same time providing fast runtimes. A point based comparison for 3D change detection is also presented by [24], who address the problem of point density variations in TLS point clouds of buildings. They propose the calculation of adaptive thresholds to determine whether changes occurred between two point clouds and consider registration errors. The tremendous advances in voxel or single point based comparison of point clouds have created many opportunities for specialized applications in change detection, including buildings. However, the comparison of changes with respect to a specific model is not yet or hardly addressed in these publications.

Construction progress monitoring: Change detection in the context of BIM and site monitoring is a common application and wide-ranging research topic. [6,39] show how change detection based on photogrammetric point clouds, occupancy grids and visibility analysis is integrated into the BIM process where it has a positive impact on project progress. The change detection method introduced by [15] is also adopted to the field of construction site monitoring. [17] transferred evidence theory and belief masses on the modeling of photogrammetric observations. They developed a method for point cloud registration as a prerequisite for geometric and semantic change detection based on voxel and occupancy grids on construction sites. [13] modeled the occupancy of space from unordered site images and analyzed the scene with a given BIM based on a robust registration approach and a proposed machine-learning scheme that automatically detects physical progress. Machine-learning and point based algorithms are also used by [35], who track building and construction progress from airborne LiDAR point clouds. Other authors focus on point clouds from terrestrial laser scanning for construction progress documentation as well. [5] contributed a method for the comparison of as-planned and as-built data based on terrestrial laser scanning. In a subsequent publication [4], the authors focus on detecting and identifying cylindrical BIM objects that are not built in their intended location and considering their completeness based on TLS. [47] presented an automated process that measures construction progress in terms of percentage of completion by using TLS data which is superimposed on the reference model. Although they demonstrate feasibility, their experiments relate only to laboratory conditions and not to real construction projects. A comprehensive overview of change detection in construction industry based on point clouds and voxel representation is given in [44].

Accuracy of TLS measurements: In this article, we refer to the measurement uncertainty of TLS. A priori knowledge of the stochastic properties of the measuring sensor of a TLS is important to detect outliers as well as to separate statistically significant deformations from the measurement uncertainty. Many authors such as [9,12,14,36,46] have comprehensively investigated error sources, influencing parameters and their effects on the geometric quality of 3D point clouds in order to allow for proper handling, compensation and prior accuracy assumption. [37,42] introduce intensity-based stochastic models while other authors particularly address systematic error models and propose calibration strategies for reducing the effect of systematic effects on practical deformation analyses [16,21,23]. All of this work provides a valuable scientific basis for evaluating TLS-only accuracy prior to using the method presented in this article.

Indoor applications and models: There are many works considering outdoor observations and the observation of sites that are still open and easily accessible in all areas because, for example, ceilings and roofs have not yet been installed. In recent years, more and more authors have taken up this issue and started to particularly address closed indoor scenarios and related point cloud applications. [29,30] addressed processing 3D point clouds especially of indoor environments for reconstruction and analysis applications. They do not examine construction site applications, but rather exploit the indoor topology of existing buildings. [20] claim that complex indoor environments still remain an open challenge for automated as-built BIM. The authors propose a method for 3D volumetric indoor reconstruction based on 3D point clouds, room segmentation and noise filtering. [45] extended existing methods and resolved the special case of indoor reconstruction of multi-room environments with curved walls based on 3D point clouds. [38] focus especially on indoor applications, because indoor point clouds are typically erroneous and incomplete. Therefore, they propose a novel method based on the combination of shape grammar and a data-driven process for procedural reconstruction of 3D indoor models from point clouds.

Uncertainty of the reference model: Uncertainty assessment is an important topic in”scan-to-BIM” applications, which is increasingly getting more and more attention. There are works such as [19,26,33] not necessarily related to BIM but with focus on reasoning with uncertain given model features. The geometric quality of a given BIM is crucial when it comes to a comparison of as-planned and as-built. The overall accuracy of this process depends not only on the measurement accuracy, but in particular significantly on the geometric quality of the given BIM. [10] address this problem and assess the impact of levels of automation and modeller training on the accuracy and precision of generated BIMs. Also [27] pick up on this with an investigation of the achievable accuracy on camera orientation based on detected line features from a given BIM, which is assumed to be uncertain.

In contrast to related work in the field of change detection for BIM, this project is characterized in particular by two major aspects: First, we address change detection primarily in indoor scenarios, because a building's interior, especially when it is under construction, is a much more challenging environment compared to its exterior. This is due to constraint scanning geometries, low observational redundancy and weak georeferencing whilst high architectonic symmetry resulting in ambiguities during processing and analysis. Most related work and approaches to change detection for BIM largely rely on outside observations. Second, we explicitly consider and integrate uncertainties using the Dempster-Shafer evidence theory. On the one hand, these uncertainties refer to the physical properties of the laser range measurement, the coordinate reference frame and the point cloud alignment but on the other hand, the novelty of the presented approach results from the consideration of the indoor model to be checked itself as statistically uncertain (Fig. 1). We derive the a priori geometric uncertainties of the BIM according to the Level of Accuracy Specification (LOA) for BIM, which is provided by the U.S. Institute of Building Documentation [40]. In contrast to previous works in the field of change detection for BIM, we are particularly capable of evaluating the significance of detected changes by considering the quality of the reference model itself.

This contribution demonstrates three main processing stages: (i) fine resolution voxel-based change detection using dense 3D point clouds from TLS (ii) modeling the occupancy of space with belief functions, which are adapted to the special conditions in indoor scenarios and (iii) reasonable assessment of model uncertainties according to the LOA specification for BIM.

i) Point clouds from terrestrial laser scanning are characterized by a generally high but also very heterogeneous point density. We use a certainly fine discretization by voxel. Integrating viewpoint information enables sort of ray tracing through voxel space and each voxel can be labeled either empty, occupied or unknown. A change detection is realized by identifying and evaluating conflicts of voxel states from different measurement epochs. Voxel size and threshold for outlier removal according to the individual point cloud characteristics should be carefully selected.

ii) In a second step, a method for modeling the occupancy of space on point-level from laser range measurements, originally presented by [15] for airborne laser scanning, is adopted to refine the results of pre-processing. The influence of a laser point on the occupancy of space at a certain location is modeled by belief functions, which in turn deliver belief masses. We demonstrate how to adjust the parameters of the belief functions in order to handle difficult ”indoor conditions” and to verify the geometric representation of a given BIM.

iii) A TLS measurement has a limited accuracy and BIM compliant building models should not be considered as error-free either. While the impact of a laser range measurement on a query position is modeled by belief functions according to evidence theory, the level of geometric accuracy for a BIM is represented in terms of classical standard deviation. We show how to merge and evaluate both uncertainty representations to infer a final accuracy assessment based on a TLS point cloud and a LOA compliant reference BIM.