Optimization of point clouds extraction for BIM applications

Abstract

BIM aims at developing information-rich, object-based models that can keep users informed throughout the implementation of project activities, including testing, operation and maintenance. Nevertheless, BIM faces several challenges, such as finding a way to retrieve quality information cost-effectively and process digital information concerning existing assets accurately. Point clouds derived from laser scanning are reality-based models that can be used in a semi-automatic manner to manage and interpret geometrical shapes of buildings with point clouds visualization. The point clouds can illustrate both complex geometry and modern constructs in regular shapes. However, it requires intensive manpower to extract and convert reality-based point data into object based-models information for the interoperability in BIM application. The existing mainstream segmentation method for point clouds data may not be suitable for the latest BIM development since it does not take into account of the considerations at the outset of a building project. Also, the complexity of building environment can result in uncertainties in object extraction quality. This study intends to propose an optimized method of “multi-level” segmentation which involves the building shape and alignment for objects extraction. This is done by extracting point clouds from reality-based data to object-based application. This method of “multi-level” segmentation which takes building shapes for objects extraction into account is proposed. The refinement of the splitting procedure is a feasible strategy for upgrading extraction quality. It is a critical criterion for sampling and estimation of object parameters from point clouds. In order to test the proposed segmentation approach, experimental measurements were undertaken in five simulated environments. It includes the scenarios of vertical shafting with traditional building approach, and precast segment placements of asymmetric bridge and flyover, indoor environment and aged buildings. Each of them was chosen with regard to the building styles and features of the building environments. The first measurement illustrates the effectiveness of the proposed multi-level segmentation for locating the defects. The second measurement demonstrates the quality of the estimated alignment and its potential to objects extraction in complicated bridge and flyover conditions. By applying findings from previous experimental measurements, the fourth and fifth measurement links up the extracted objects’ parameters to form the object-based information of an old building and indoor environment which supports quantities estimation. The test results deduced from the prototyping of the proposed workflow indicate that the extracted building object quality has been improved by the proposed workflow, in comparison with the existing ones. This improvement is even more significant where the construction conditions are complex. This optimization translates the massive data storage reality-based point clouds to the object-based BIM information with increased interoperability within the AEC community.