4D Point Cloud Video Processing for Digital Twin Construction

4D Point Cloud Video Processing for Digital Twin Construction
Project ID: 2627Arc1001
Research Mentor: Prof. Fan Xue
Contact Person: Prof. Fan Xue
ABSTRACT

Traditional 2D CCTV struggles to capture the geometry, occlusion, and motion inherent to active construction sites, especially in low light. Time-dynamic 4D point clouds (4DPC), i.e., sequential LiDAR frames encoding x, y, z coordinates and time, provide a low-cost, centimeter-accurate spatiotemporal record of site elements including cranes, workers, and components. This project introduces undergraduates to 4DPC processing for construction digital twins, focusing on three areas: (i) compression and streaming of large LiDAR sequences, (ii) detection and tracking of moving objects, and (iii) registration of observed motions to as-designed models for site monitoring, safety alerting, and progress tracking.

Each student will address one of the following research questions:

(i) How can 4DPC data be compressed, indexed, and replayed — comparing, for example, depth-image and H.264-style coding with lossless archival methods — while preserving construction-critical geometry?

(ii) How can moving vehicles, modules, and workers be segmented and tracked across frames under limited labeling conditions, leveraging geometric and intensity features alongside optional BIM reference objects?

(iii) How can frame-wise 4DPC be registered to as-designed BIM such that hoisting operations and site activities are represented as time-stamped digital-twin trajectories?

Each study comprises a literature review, algorithm prototyping, and evaluation using existing 4DPC datasets supplemented, where permitted, by laboratory or site scans. The principal outputs are a research poster and a short methods paper.

SKILLS & EXPERIENCE REQUIRED

Preferred Background

Surveying, civil/geomatics engineering, computer science, AI, data science, or a related STEM discipline.

Technical Skills

  • Python programming (NumPy, Open3D, or similar); Pytorch/C++ is helpful but not required.
  • Foundational knowledge of machine learning or computer vision (e.g., classification, clustering, tracking).
  • Familiarity with 3D data representations such as point clouds, meshes, or BIM/GIS viewers.
  • Willingness to learn LiDAR data formats (.las, .ply, .lvx) and basic data processing pipelines.

No prior site experience is required; fieldwork will be undertaken only where arranged and subject to safety induction.

Sign In

Register

Reset Password

Please enter your username or email address, you will receive a link to create a new password via email.