Aerial Path Online Planning for Urban Scene Updation

TL;DR

This work reframes urban scene updates as an aerial path planning problem that targets change detection rather than full-scene reassembly. It introduces a two-path framework— a prior path guided by static priors and a real-time path that activates upon detecting changes—driven by a changeability heuristic that leverages prior reconstructions ($T_1$) and statistics (e.g., from the $WUSU$ dataset). By combining candidate-view generation on a safe-height plane with a probabilistic change model, the method efficiently identifies and fully explores change areas, producing convex hulls that feed downstream reconstruction. Empirical results on real-world UrbanBIS scenes demonstrate significant reductions in flight time and computational overhead while maintaining update quality comparable to full re-exploration, enabling scalable, adaptive UAV-based scene updates in complex urban environments.

Abstract

We present the first scene-update aerial path planning algorithm specifically designed for detecting and updating change areas in urban environments. While existing methods for large-scale 3D urban scene reconstruction focus on achieving high accuracy and completeness, they are inefficient for scenarios requiring periodic updates, as they often re-explore and reconstruct entire scenes, wasting significant time and resources on unchanged areas. To address this limitation, our method leverages prior reconstructions and change probability statistics to guide UAVs in detecting and focusing on areas likely to have changed. Our approach introduces a novel changeability heuristic to evaluate the likelihood of changes, driving the planning of two flight paths: a prior path informed by static priors and a dynamic real-time path that adapts to newly detected changes. The framework integrates surface sampling and candidate view generation strategies, ensuring efficient coverage of change areas with minimal redundancy. Extensive experiments on real-world urban datasets demonstrate that our method significantly reduces flight time and computational overhead, while maintaining high-quality updates comparable to full-scene re-exploration and reconstruction. These contributions pave the way for efficient, scalable, and adaptive UAV-based scene updates in complex urban environments.

Figures (12)

• Figure 1: The input to our method is a labeled reconstructed model of $T_1$, where each semantic label is associated with a prior probability (Sec. \ref{['sec:changeability']}). We first plan the prior path by minimizing redundancy while maximizing potential changeability and coverage (Sec. \ref{['sec:prior_path']}). As the drone flies along the prior path and detects a change area, the real-time path planning (Sec. \ref{['sec:realtime_path']}) is triggered to explore the entire target change area by analyzing the diff images. The output of our system is a set of convex hulls representing all detected change areas.
• Figure 2: An example of our real-time path (detailed in Sec. \ref{['sec:realtime_path']}) applied to a scene from UrbanBIS yang2023urbanbis. The $T_1$ reconstructed model (top left) serves as the prior, showing three existing buildings, while the $T_2$ model (bottom left) serves as the ground truth, where three buildings have been removed. As the drone detects changes in the scene, it dynamically adjusts its path by identifying differences in captured 2D images (middle). Subsequently, the detected point cloud (bottom right) and the convex hull of the change area (top right) are generated based on the updated real-time path.
• Figure 3: Visual comparison of the efficiency of our method (b) versus zhou2020offsite (a). The experimental setup is detailed in Sec. \ref{['sec:exp_efficiency']}, and the corresponding statistics are presented in Table \ref{['tab:exp_efficiency']}. Our method demonstrates significantly improved efficiency, making it well-suited for downstream tasks such as reconstruction.
• Figure 4: Qualitative comparison of path planning methods for detecting and extracting change areas. Refer to Sec. \ref{['sec:exp_change_exp']} for experimental details. The Region-Division method liu2021aerial is evaluated with two grid sizes: $\frac{1}{2} \epsilon$ (a) and $\frac{1}{3} \epsilon$ (b). Our approach demonstrates accurate detection of change areas while requiring fewer viewpoints and shorter trajectories, with quantitative metrics provided in Table \ref{['tab:exp_change_exp']}.
• Figure 5: Visual comparison of reconstructed buildings using either the coarse proxy approach from zhou2020offsite or the convex hulls generated by our method. Quantitative statistics corresponding to these results are provided in Table \ref{['tab:exp_recon']}. Reconstructions are shown for two different scenes, we provide more visual and quantitative comparisons in the Supplementary Material.