Reducing spatial error in mobile laser scanning by real-time uncertainty visualization and human-machine interaction

Abstract

Scanning is a key element for many applications in the AECO industry. It provides point clouds used for construction quality assurance, scan-to-BIM workflows and construction surveys. However, data acquisition using static laser scanners or photogrammetry methods is lengthy and requires even lengthier subsequent processing. A quick and apparent escape from this problem might be mobile mapping solutions mainly based on lidars. However, current hand-held scanners suffer from drift, skewing point clouds and thus, increasing their spatial error. In this paper, we present a novel, real-time and fully explainable method exploiting human-machine interaction to increase the correctness of produced point clouds. Our method progressively reconstructs the scanned scene and predicts the regions of a potentially high error with a 95% confidence level. The user can then revisit these parts of the scene, which adds additional constraints on the underlying probabilistic graphical model, thus reducing the drift and increasing the confidence in the correctness of these regions. We build a prototypic lidar-based mobile scanner, implement our method and test it in a case study. The results show that the areas identified with a relatively high spatial error indeed suffer from it, while predicted areas with relatively higher correctness do have a smaller spatial error.