Active UAV Re-Scan Planning for Dynamic Object Artifact-Free Point Cloud Mapping

The research aims to design and evaluate an uncertainty-aware UAV mapping pipeline that: 1. Detects and quantifies dynamic object artifacts in UAV point clouds. 2. Estimates where re-scans are most beneficial. 3. Plans and executes UAV re-flights (active mapping) to minimize ghost points while respecting flight-time and battery constraints. 4. Demonstrates improved map quality and efficiency compared to full re-flights. The study will begin with UAV data collection over environments containing dynamic elements such as vehicles, pedestrians, or vegetation. Point clouds will be generated from either UAV LiDAR or photogrammetry. Dynamic object artifacts will first be detected using robotics principles including occupancy grids, ray tracing, and temporal consistency checks. To enhance detection, a lightweight deep learning model will be trained to predict the likelihood and uncertainty of artifacts from local point cloud patches, optionally supported by image features. Based on the uncertainty maps, an active re-scan strategy will be designed, where Next-Best-View or coverage optimization techniques identify the most beneficial re-flight locations under battery and time constraints. The UAV will then execute these targeted re-flights to replace unreliable data with new static observations. Finally, the improved point clouds and derived DEMs/DSMs will be evaluated against baselines such as single-pass mapping, full re-flights, and inpainting-based post-processing.