An Automated Target-Oriented Scanning System for Infrastructure Applications

Abstract

Point cloud pre-processing is essential for emerging applications such as digital twinning but currently requires a lot of manual effort before the resulting data can be used. Practitioners usually use default scan range settings to take full scans, which generate huge point cloud datasets containing millions of points. However, only a fraction of the dataset is used for subsequent twinning processes and the remaining data is “noise”. Researchers need to perform substantial cropping work to enable the point cloud can be used for detecting objects of interest. However, the problem of object detection in the post-processing stage also remains unresolved. This paper describes a new system TOSS to conduct a target-oriented scanning process. It streamlines the scan-to-gDT (geometric digital twin) process by automatically identifying the region of interest and its corresponding scanning path. TOSS consists of a cost-effective 3-DoF rotational laser scanner, a vision-based object detection algorithm, and a geometric-camera-model-based scanning control algorithm. Preliminary results on a real-world bridge indicate that TOSS can produce accurate scans of regions of interest (average: 95.5% Precision and 89.4% Recall). It is fully scalable and can be adapted to various infrastructure types, including buildings, bridges, industrial plants, tunnels, and roads. The algorithms also have great potential to be embedded in a traditional scanner’s software.