Adaptive grid-based decomposition for UAV-based coverage path planning in maritime search and rescue

TL;DR

The paper addresses efficient coverage of polygonal search areas by UAVs in maritime SAR through an Adaptive Grid-based Decomposition (AGD) that reduces the number of grid cells, paired with a Mixed-Integer Programming (MIP) model to produce a time-minimizing coverage path. AGD tunes grid cells against the UAV camera footprint, using a decomposition channel between $y_b$ and $y_t$, and computes adjustments via $l$, $n$, $e$, and $\Delta$ to fit the footprint. Experimental results on multiple polygon cases show up to $20\%$ savings in coverage time and up to 12 fewer cells compared to standard grids, demonstrating meaningful improvements for SAR operations. The approach provides a practical, scalable framework for single-UAV CPP and suggests extensions to multi-UAV scenarios and broader path-planning applications.

Abstract

Unmanned aerial vehicles (UAVs) are increasingly utilized in search and rescue (SAR) operations to enhance efficiency by enabling rescue teams to cover large search areas in a shorter time. Reducing coverage time directly increases the likelihood of finding the target quickly, thereby improving the chances of a successful SAR operation. In this context, UAVs require path planning to determine the optimal flight path that fully covers the search area in the least amount of time. A common approach involves decomposing the search area into a grid, where the UAV must visit all cells to achieve complete coverage. In this paper, we propose an Adaptive Grid-based Decomposition (AGD) algorithm that efficiently partitions polygonal search areas into grids with fewer cells. Additionally, we utilize a Mixed-Integer Programming (MIP) model, compatible with the AGD algorithm, to determine a flight path that ensures complete cell coverage while minimizing overall coverage time. Experimental results highlight the efficiency of the AGD algorithm in reducing coverage time (by up to 20%) across various scenarios.

Figures (5)

• Figure 1: (a) The UAV's camera footprint; (b) By considering $r$ as the radius of the camera's footprint, (I) shows the grid cell dimensions, and (II) shows the grid cell dimensions after a $\Delta$ adjustment to one edge (subtracting $\Delta$ from one edge).
• Figure 2: An illustrative example demonstrating the step-by-step implementation of the AGD algorithm. The polygon vertices are $\{(0, 0), (10, 0), (12, 5), (8, 8.5), (2, 8.5)\}$, and the radius of the UAV's camera footprint ($r$) is $\sqrt{2}$. In each iteration, the gray area represents the decomposition channel between $y_b$ and $y_t$. Note that in the standard grid-based decomposition of the search area all cells' edge size is $\sqrt{2}r=2$ (see Fig. \ref{['fig:footprint&grid']}b case (I)).
• Figure 3: Cases in Table \ref{['tab:case_result']}.
• Figure 4: Cases in Table \ref{['tab:case_result']} after the AGD algorithm implementation.
• Figure 5: Example 1 in Appendix. (a) The main search area; (b) after applying the $A$ matrix (30-degree clockwise rotation); (c) after applying the $b$ vector (-4 shift in the Y-coordinate); (d) after applying the AGD algorithm; (e) after applying the $-b$ vector (+4 shift in the Y-coordinate); (f) after applying the $A^{-1}$ matrix (30-degree counterclockwise rotation to restore the main search area with the AGD decomposition).