SemReg: Semantics-Constrained Point Cloud Registration
SemReg incorporates semantic labels as geometric registration constraints, enabling robust point cloud alignment even in textureless or symmetry-ambiguous regions where purely geometric methods fail, with state-of-the-art results on indoor and outdoor benchmarks.
Sep 1, 2024
TopFormer: Topology-Aware Transformer for Point Cloud Registration
TopFormer augments transformer-based point cloud registration with topological structure descriptors, improving correspondence quality in geometrically repetitive and low-texture regions by encoding how local surface neighbourhoods connect globally.
Apr 1, 2024
WTAYOLO: A Detection Model for Weak Scratches on Metal Surfaces
This work focuses on wtayolo: a detection model for weak scratches on metal surfaces, contributing to Wei Pan's research thread in 3D vision, optical metrology, geometry processing, and industrial inspection.
Jan 1, 2024
Weighted Point Cloud Normal Estimation
A weighted normal estimation scheme adaptively weights neighbouring points based on local surface geometry confidence, producing more accurate normals at sharp features and noisy regions than standard PCA-based and learning-based methods.
Jul 1, 2023
3D reconstruction of moving object by double sampling based on phase shifting profilometry
A double-sampling strategy in phase-shifting profilometry acquires two temporally interleaved fringe sets to estimate and compensate motion between exposures, enabling 3D reconstruction of moving objects without specialised high-speed hardware.
Jan 1, 2023
Parallel Curvature Filter for High Performance Image Processing
GPU-parallel implementation of curvature filtering for image smoothing and denoising, achieving 33.2 Giga pixels per second throughput—orders of magnitude faster than the CPU baseline.
Jan 1, 2019
Design of LEGO Assembly Transformation
A design method for reassembling an existing LEGO model into a new target shape, laying the algorithmic groundwork for the full transformation pipeline.
Jan 1, 2017
Moving Sequence Generation Based on Hungarian Method for Pick-and-Place Process
Optimal moving sequence generation for robotic pick-and-place assembly using the Hungarian assignment algorithm, minimizing travel cost in automated 3D construction.
Jan 1, 2015